I at least didn't see this. I had a baseline evaluation from before Nov 3 on Opus 5.5 medium, so I re-ran it this morning.
Eval context: 30 bug fixes and features evaluated by tests the agent didn't see. Ran on Opus 5.5 at medium effort in Claude Code.
20 of the tasks were last run on Nov 1, twice each. Nov 1 solved 36/40, and today solved 19/20. Thinking tokens per task were 0.95x Nov 1’s and output tokens 0.97x. The one significant change was speed: median time to first token decreased from 3.0s to 1.7s.
Caveats: one run per task today, medium effort only.
from my basic stats knowledge, it's not reliable to calculate the confidence interval of the sample proportion with fewer than about 10 samples on both sides. So if you're only seeing 1 or 2 failures in 20, and that approximately held as you scaled, you would have to run about 200 tests with each in order to get at least 10 failures in order to calculate the confidence intervals of the two samples and see if they overlap or not.
"Nerf tracking" is adversarial and exploratory. It's not something Anthropic wants anyone to do, and it's not something that the "nerf tracking" designers are necessarily doing in a foolproof way. Don't assume too much from them either way.
Sometimes you just have insufficient evidence to be sure of something either way, and nobody gets to win the argument or shut somebody else down.
In this case, some have reported analytical work[0] that goes beyond "gut feel" anecodotal evidence, but even that's not going to conclusive and may not be trustworthy at all.
These nerf trackers do not work. I wish people would stop posting them.
They don't detect a quantized model until the quantization gets down to 4 bits or less.
Humans absolutely do detect quantization well before 4 bits...
Try it yourself if you want. Try a small local model, both quantized and not, and I would be very surprised if it isn't immediately obvious which is which.
Its incredibly easy to roll out a compute instance to X% of users, or route traffic between compute instances
so not everyone will have the same experience. If I was saving on costs that would be the easiest way to do it without deploying, without any nefarious intent. The persons in charge of that likely don't even know about how individual users are impacted
At what point does this become false advertising, or even fraud? Imagine paying AWS for a high-performance server, but 5% of customers who order one quietly get a slower, cheaper server and still pay the full price. That's a serious claim and makes me skeptical Anthropic is doing something like that without more evidence.
It would be nice to know how they select. I max out my $200 subscription every week and I was selected. On Saturday, the Opus on my account that happened to be supervising a running live process, became unusable.
It would be trivial for them to opt-out accounts that are doing benchmarking from a planned regression.
I really really really do not trust companies on the verge of IPOing from goosing their numbers. Even if the lawyers are up their ass, the social pressures are more than enough to compel the behavior. For every good person willing to do the right thing even at the cost of a delayed career progression, there are 20 people willing to do anything not even for a dollar but for an "atta boy!" from the boss.
That's a major reason to use local models. Or open weights models on a trusted cloud provider. Even if they are way less capable, they're more predictable
I used to be skeptical of nerfing, but as of a week or so ago I've noticed a severe degradation of quality for GPT Astra. It struggled with very simple tasks, and it often failed to follow my instructions.
I hear a lot of accusations of nerfing as in, saying that the model itself has been quantized.
I doubt this is the case. I suspect all variation in quality and output can be explained in terms of prompt differences, and changing how much thinking happens at each level, as we see in the post above.
Suppose a company has a limited amount of compute. What do they then do if peak demand exceeds what they can provide? They don't turn off people's access...
And what about A/B testing? Anthropic doesn't even try to hide that they do this. Limited compute and A/B testing alone, would guarantee a nerfed experience some of the time. That is, you don't even need a nefarious reason.
It's been the same for every version the last year.
Oh, this Opus X.Y is incredible.
Hmm... Now it's dumb.
Rinse, repeat.
Doesn't really matter if the cause is quantization, system prompt changes, changes to thinking etc.
It's hard to see how continuously rug pulling users at every release is a good idea. Only reason I can fathom is that it's needed to keep costs inline.
Either that or it's some weird tweaking to the prompts to fix whatever. But given the prevalent sentiment that it's worse that sounds like a long shot
it's easy to suspect the hosting companies trying to do what they can to reduce cost-to-host up until the very threshold of perceptibility, so it's an easy guess to say they're probably doing something that shifts performance up or down between version releases..
but realistically I think you're right, it's prompt quality and the inherent non-determinism elements of LLMs showing up; when both of those things meet you get brain dead outputs.
Everyone likes to shout "slot machines! non-determinism!" until it actually explains something, heh.
YES. To the point that I literally stopped using it, and have moved BACK to Claud.
This is getting ridiculous. The number of times I have switched from one model back to the other is crazy. I do not have any proof, so its quite possible I'm just delusional. AI psychosis? Either way, it FEELS like these models work really well for a while and then fall off. Switching to the latest model on the competitor always seems better, until it too seems to taper off. By then, a new model is usually available at the competitor, so I switch.
Its possible there's no degradation at all and just a PERCEPTION of it. But to me the end user, it really doesn't matter.
People are prone to seeing patterns in noise especially after suggestion. It's not hard to imagine someone concluding the model is nerfd after few bad responses in a row
It could apply to me but I run evals every time someone at work complains about this and it’s never shown any more or less variance than regular usage. I don’t have statistical relevance with my evals but I do have a decent history with them.
People seem to underestimate how high normal variance is with these tools.
i'm not parent, but to answer your snark : there are efforts a person can take to try to reduce things like suggestability in ones' own personality. Same as anything, the brain can be trained, thus it should go without saying that some of us react differently in the face of suggestion -- both for the better and the worse.
idk, it hasn't felt nerfed at all to me, and it's not like i'm just doing trivial CRUD webapps either. opus 5.5 is very successfully porting my phone to postmarketos. claude just got the touchscreen up and working (and made a little program for me to test it, and it's legit).
If this turns out to be accurate, I think that would be a pretty unethical thing for Anthropic. Even if it is just a "we had to allocate compute to this new model release".
It should be fraud to swap out the model. We need a term for this, something like a bait-and-switch.
I'm pretty convinced that those who have constantly been posting these complaints, probably for every single model release out there, are indeed wrong.
I believe fraud is a sufficient term. You claim to be selling X but you actually sold them Y. Bait and switch is a selling tactic to avoid fraud where they make you believe product is X and then when you go to buy they tell you X isnt available but they have Y. While also unethical, it is not fraud because they did tell you what it is you are buying before you buy it.
if they just straight up dont sell you something you bought then its fraud.
I think there is a ripe opportunity for expanding the definition of fraud to include changing the "ingredients" of a named product on the market. It is not limited here, we see it with food, and even in the medical field
(I will die on the hill that medical fraud was committed when mRNA therapy was used as a treatment for covid and it was 'sold' to the public as a "vaccine" despite not being the same thing as any vaccine before it. The definition of vaccine was expanded to include it without sifficient public disclosure to warrant informed consent imo. This is not a statement about conspiracy or the efficacy of vaccines or a moral judgement on whether it was for the greater good or not.. it was simply an act of fraud that robbed the public of proper informed consent by making people believe it was X when it was really Y.)
So yes, I think there is a large need for the public to at the very least put a strong handle on the ability to change what a product name means. It is a trivial problem to solve considering it is only a problem due fraud or fraud adjacent activities.
heres the solution: You have your product name, and then any changes to it REQUIRE a new version name. Like how you have Coke and Diet Coke, only formalized such that Coke cannot legally replace the ingredients of regular Coke with the ingredients of Diet coke and still call it Coke. With Software they could at least use version numbers and release modifiers. most of the time its built in already, but it needs to include the full stack. if they change anything that has an interaction with the user it should change the label seen by the user for the product.
I don't have a formal benchmark, but daily work (coding mostly, plus some complex architecture design and review) has had consistent outputs today for me. No noticeable nerf from my perspective. I definitely took notice in some past instances, but if they nerfed Opus 5.5, then I guess it was just so good that I was nowhere near its limits.
In a normal world you are not allowed to degrade the service during your agreed and paid contract. You would get sued. Apparently this is perfectly fine in this current world.
Benchmarks don't do a good job at capturing the difference in quality of frontier models over smaller models.
Opus 5.5 was so good that if it wasn't nerfed - at least just looking at the benchmarks - the difference in cost here does not seem to justify using this for anything besides design and review.
That's like maybe 1% of total token usage for SWE BUT a better design can have enormous benefits, so it's almost always worth using the best model for that.
yeah, we need to make sure we don't get stuck with these two evil corps. Dario is really the worst. Plays the drama queen but he is very hard in all things dark(i.e. dark patterns).
Interesting - I had the same feeling but mostly with 4.6 when I "downgraded" for some easier/trivial tasks: The model forget things I posted 20 min ago in the chat (and the same file was in the context as well)
I'm not surprised at all that they might be doing shady shit, my free request was used up in less than 7 minutes and said pay to continue, no thanks. I've paid for monthly before, never again I don't care how good the model is. I asked about the extreme limited usage in their discord and got banned, not a good look. You don't have a way to measure what you're actually paying for and they can change it at any time.
With early handouts of Qwen4 starting to be reported 5 days ago ... so if they're good, they would have had some hours to evaluate Qwen4 against Claude 5.5 and, then or right after was when Anthropic decided to put out 5.2 ...
So yeah ... benchmaxxing and all that. BUT nevertheless, the published results seem to say that Qwen4-Max appears to match Opus 5.5 on some benchmarks, and Qwen4-27B (probably by going even harder than Qwen3.8-27B on thinking/inference time compute) can beat Sonnet 5.5 on some things.
I guess they're not taking any chances. Keep in mind that Anthropic has declared they have non-public models (it's in their IPO filing). These are referred to by the shorthand names Model 2, Model 3, and Model 4. Essentially 4 is better than 3 is better than 2.
Fable 5.2 is "2.5" so to speak. So Anthropic was holding back, but not anymore. And now they can dial the performance up once more with what they have ready, and probably once more after that.
> Our first formalization of Fermat’s Last Theorem, performed by an unreleased research version of Claude, took 11 days. It relied on a purpose-built harness along with guidance from a human researcher. By contrast, a team of 1,000 Fable 5.2 agents fully formalized the theorem in just 30 hours, with minimal human intervention. Fable 5.2 started from minimal prompts, built its own ways to coordinate, and used no mathematical references, only a list of intermediate theorem statements to score progress. It needed one-third of the action tokens as our first formalization and produced a proof seven times shorter. As can be seen in the figure below, Opus 5.5 (with the same 1,000-agent setup) completed only 26% of the formalization in 18 hours, and barely made any more progress after 30.
Most people don't really have intuitions on what it means to push the frontier now (because AI has surpassed everything they are good at, for example, coding). We are truly inside the singularity (using Kurzweil's definition, slightly paraphrased - the time when technology progresses so fast no one can keep track with it).
> Fable 5.2 started from minimal prompts, built its own ways to coordinate, and used no mathematical references, only a list of intermediate theorem statements to score progress. It needed one-third of the action tokens as our first formalization and produced a proof seven times shorter.
Is this really so astonishing, assuming that the 5.2 training data certainly had the 'traces' from the Fable original run in it's training data.
I feel like I'm increasingly reading AI current events is more like learning about historical events. By the time information has entered my head, there is even newer information out there that already makes what my brain is only currently actively processing, obsolete.
I have to imagine if I worked for one of these companies on the frontier models it would be very difficult to leave for that reason. Thee pay is also pretty good.
Yes, and not very helpful without knowing how many agents were used by the custom-built harness. The 30 hrs vs. 11 days means nothing without at least some token-to-token comparison. Without that, they are pumping up their new model with half data - better to just say "we solved it in 30 hours" and who cares about your internal, custom model.
All the human engineers will be in harnesses in the future.
Within three years most software developer jobs will consist of taking instruction from the AI systems. The humans won't see it that way, but that's what it will actually be. Realistically the thing that will happen is the humans will be put into harnesses (cubicles) and prompted by the AI (I need you to plug that server in; I need you to give me access to such and such; now do this, then do that). Most tech jobs will consist of AI telling the human what the AI needs the human to do next, de facto the human will be in a harness and will be getting told what to do.
If you take Daniel Schmachtenberger's thesis, these models are the latest extension of an ongoing misaligned moloch-like entity, using our institutions/economy/corporations, etc. as its organelles. All powered, most recently, by fossil fuel.
The ownership class, the capital class - whatever people prefer to call it, will still call all the shots as they always have, and always will.
So whether that's a CEO at the top instructing a human manager that then instructs AI systems, which then put into motion plans that include the harnessed humans getting prompted, or however it's decided to best configure the top operational tiers - that's all fairly irrelevant ultimately. All human labor, and all AI labor, will be in harnesses. Realistically all human labor has always been in a harness.
The people in charge will be the people that control the capital. They will instruct the AI to their bidding, they will control the data centers the AI lives in, they will own the robots doing the labor in factories, they will own the factories, they will control the grid, they will control the electricity generation, et al.
It's true now, it'll be true tomorrow. Nothing fundamental will shift at that tier: don't be labor, own.
If you're a software developer, you're increasingly depreciating labor. Every unit of improvement the AI systems see, the average human software developer's labor value diminishes. It's diminishing extraordinarily fast, kind of like a confrontation between increasingly productive machines and physical human labor during the industrial revolution. You might as well picture a human with a hand plow taking on a modern industrial tractor.
> We're all going to be harness engineers in the future
No, that will be the model's job. We will largely be unemployed. Of course, we can surely depend on the largesse of Altman and Amodei to keep us fed and sheltered.
Isn't this part of the lore behind the Dune universe?
Like, we're dropped into a post post post singularity world where AI already did all of that and is banned? as kind of a footnote that's never further explored in movie renditions that have been released
Dune books by Frank Herbert never actually went into what "Butlerian Jihad" was.
The semi-canonical Dune Encyclopedia (disregarded by Frank's son) differs vastly from prequels and generally points to it being a self-inflicted wound on humanity that opened the doors to widespread return of feudalism and essentially stopped humanity for a 10,000 years
> We are truly inside the singularity (using Kurzweil's definition, slightly paraphrased - the time when technology progresses so fast no one can keep track with it).
I gotta admit, I am jelly as fuck I don't have access to a billion dollar datacenter. Would LOVE to put 1000 fable 5.2 agents to self-coordinate and get them to do something for me. Fuck life is unfair.
I'm using Opus 5.5 and I cant help but think that I might very well be one of the few people who do so in my local area ( developing country, one of the poorest province ).
Had Opus make software for some local business and seeing it blew people's mind.
20$ aint cheap for us but I'm actually considering upgrading to the 100$ tier.
Am just jelly people are using 100-200$ subscription ^_^
I’d be happy even without having access to a billion dollar datacenter, if I could use local models properly, but even that requires some serious hardware that is out of my reach. We live in weird times: I have lots of hope for the future, but meanwhile AI access is pretty iniquitous indeed.
Sub 1% of the planet has access to cutting edge frontier AI.
Sub 1% of the planet has any practical use for the top tiers of AI systems. That will likely always be true.
The gap between the haves and have nots will drastically widen further, and very quickly. The bottom ~95-97% will be left behind in a way that will be impossible to overcome. The age of AI + robotics will just make it that much harder to climb for most people, these are systems that are in aggressive, direct competition with human labor. And in short order the AI + robots will be vastly superior to human labor (it doesn't really matter if we're speaking in terms of 10 years or 30 years).
AI will consolidate the winnings. The gigantic gains by the elite rich related to AI over the past few years, is merely a hint of what's to come. Capital will win as it always does.
There will be some gains for most everybody else, those gains will be comparatively trivial though. A house bot that saves people an hour of labor here and there, AI that automates some trivial life tasks.
AI has no money, it has no wealth, it owns no capital. For now. If you want to avoid the brutal human vs AI/robot competition arena: own capital, own the AI, own the robots.
I’ve had a chance to use fable 5.2. Some observations:
- Anthropic is finally well balanced across models. Meaning: Opus 5.5 is so good that I do not feel Fable FOMO as I did when it was Fable 5/5.1 vs Opus 4.8/5. And Sonnet 5.5 sometimes makes me forget about Opus.
- Fable is good, but you really have to push the ambition of your prompt in order to see proper separation from Opus 5.5. I personally have a hard time telling the difference between Fable 5.2 and Opus 5.5 for ordinary coding tasks.
- It’s way faster and more reliable: one of the worst things about fable 5.1 was how slow it was an, on top of that, how frequently it would suffer from max output token overflows. I got none of these issues with 5.2.
- Fable is exceptional at drawing. It’s ability to render an image with JavaScript actually has me questioning the value of dedicated image models. I prefer the images I’ve gotten fable/opus to make for me over those from gpt-image, it’s wild to think we’re at this point.
> Fable is good, but you really have to push the ambition of your prompt in order to see proper separation from Opus 5.5.
I love seeing this verbalized. We are pretty much at the point that increased model capability will only have utility to folks that are willing to let go of even more understanding of whatever the model is producing.
The best models can/will produce a year's worth of output that would take several days to weeks to properly understand in the span of an afternoon.
For everyone that still wants to feel like they own the end result, just use the lesser models with less ambitious tasks.
Caveat: I think this mainly applies to coding. I would still like the best models possible to be working on e.g. new drug research to cure our various ailments.
> Fable is exceptional at drawing. It’s ability to render an image with JavaScript actually has me questioning the value of dedicated image models
I’ve been shifting away from using more traditional GenAI models in favor of just giving LLMs access to a custom drawing language (LOGO-like) to generate imagery and its been going shockingly well - Opus 5.5 in particular.
I put together a simple LLM driven skill that creates old adventure game (EGA) environments from simple pencil sketches and it works really well. And being procedural, they're much easier to make surgical changes to without degradation.
Does Fable 5.2 outperform Opus for data science work?
Not just "iterate on this XGBoost model" but actually looking at a problem and designing and carrying out a short-term R&D program including probability modeling, model fitting diagnostics, how to respond to various kinds of results, how to interpret/reason about what it finds, etc. That's really where I saw all the models perform poorly, and where I saw the first big step increase from Opus 5 (pre-5.5) to Fable. But then Opus 5.5 came out and I haven't needed Fable since.
Mind you, I'm doing slightly different work (and I'm very much "in the loop" still) than when I performed that side-by-side test. But it's really amazing what Opus 5.5 can do. I now have a powerful data science AI assistant that not only can write any script I need it to (including some scripts I didn't even realize I needed), and it has an encyclopedic knowledge of everything ever, and it constantly checks me on oversights, it handily derives expressions and clever model reparameterizations that aid, it has great insight into model behavior, and it has an amazing ability come up with ad-hoc experiments and tests that would have taken me 2x as long to even think of (let alone implement by hand).
It's not at the scared-to-lose-my-job phase yet, but it's become a tremendous force multiplier for a smaller data team.
I'm teaching "Data Science" to computer science majors next spring, and I'm overwhelmed by the prospect of what to teach them in the age of AI. It seems like Fable already way outperforms anything I could historically expect my students to do at the end of the semester, and perhaps even outperforms me. I can describe a vague analysis on the data, and it will handle every step: transformation, preliminary analysis, formulating more precise questions, creating visualizations, running statistical tests, interpreting results, thinking about stakeholders... And in some of those steps, it finds things that I might not have considered.
I wish I had a bigger N so I could rely on that experience a bit more, but it sure does feel like I'm not really contributing anything in terms of "Data Science expertise" in my own research right now.
And none of it matters ... because just as soon as you are happy with things "balanced across models" Anthropic will nerf the models you were happy with, and force you to use new and more expensive ones.
> - Why are they stacking Fable xHigh against Carbon Med?
Because on that specific benchmark (FrontierCode 2.0), Medium is Carbon's best effort level and xhigh is Fable 5.2's best effort level. You can see that on the graph your next point is about.
And it's just that one benchmark that's at medium. "All scores show each model’s highest-available effort unless otherwise noted".
> - How did they produce such a bizarre effort-intel curve for Carbon that goes against every available benchmark?
The graph is labeled as being about FrontierCode 2.0. Those appear to be Carbon's scores on the official FrontierCode 2.0 leaderboard: https://cognition.com/blog/frontier-code-2.0
> - Why did they leave Carbon-- positioned primarily as a cyber model-- off of the single cyber benchmark comparison?
> - Why do they switch to comparing against Argon (rather than carbon) for robodojo?
Because no numbers are available for Carbon for those benchmarks, and the model isn't publicly available so they couldn't run the benchmarks themselves?
And from your followup:
> where's Astra on that CVE chart
To run Astra on a vulnerability finding eval, you'd need to be part of OpenAI's verified cybersecurity program. Otherwise the results would be tainted by refusals. Do you expect that OpenAI would grant such access in this case?
> Because on that specific benchmark (FrontierCode 2.0), Medium is Carbon's best effort level
I am referring to the chart that cross-compares multiple benchmarks, which Artificial Analysis clearly shows Carbon-high as superior on. Further, AA's FrontierCode chart places Carbon-high as having a higher score on Frontier 2.0 than Carbon-med (https://artificialanalysis.ai/evaluations/frontiercode-2-0?m...)
Where it gets very interesting is when you look at the Cognition "FrontierCode" publish date (2026.10.27), and the revision as of 10.28 in web archive (https://web.archive.org/web/20261028175237/https://cognition...). This revision has a VERY different Carbon effort / intelligence curve where high beats medium.
It is only with the latest revision, as of today, that all of a sudden Carbon's benchmark curve goes cone-shaped. Curious, indeed....
Arena.ai released carbon tests 12 days ago - the only service that I have seen on yt that gained access. And you can compare to other SOTA models in their video.
Just to follow-up-- artificial Analysis shows no justification for that Carbon effort curve, but it does show Carbon on medium smashing Fable on things like Terminal-Bench and AA-Omniscience v1.1 Non-Hallucination Rate.
But I suppose the reality is rather obvious when you note the funny games they're playing with the graphs, and the things they won't talk about (where's Astra on that CVE chart?)
Setting aside the comments about nerfing, do normal mortals out there have workflows that really require Fable/Mythos and don't just work fine on Opus 5.5?
I've been using Opus 5.5 on xhigh for some decently aggressive software development on a ~125k-line project, and it has been fine for me. I'm not really willing to dive into a "rewrite the whole project, don't make any mistakes" level of refactoring. I'm not looking to try to prove Fermat's Last Theorem myself. Is there anything that I'm truly missing?
I’ve found that Fable is good when the model intuition really matters. For instance, picking model architecture in a machine learning problem. Opus will grind the implementation just fine, but having intuition for picking a good research direction is better with fable
I am in the exact same boat as you. Opus 5.5 on High solves all engineering problems i encounter in my SWE job. If it doesnt do well, its nearly always requirements engineering: I did forget to specify something or there was a misunderstanding. Or taste approach wrt to architectures chosen. None of these are really different if I use a stronger model.
I am also happy with this current state, as I believe it will at some point allow me to transition to open models once they reach the level of opus 5.5. high. But its probably futile to try predict the future in these times.
Eg. https://tfmbot.com is an implementation of a popular board game. Opus and even sonnet handle that part. Each week i spend my remaining credits on fable to think how to improve the ai opponents. It plays at a very high level (can get into the top 100 on BGA if i let it). Opus alone couldn’t get it to the level it’s reached.
If you understand your problem well and know 80-90% what you want it to look like (or it just isn't very hard), it's just more cost effective to use a smaller model. Even Opus is overkill for a lot of the work I do. Most code I produce is from Luna Max implementing against fairly prescriptive plans made by bigger models.
Yeah, I just threw it into a refactoring problem, and it started to produce a mess because of a bad API I decision I made 5 years ago (when I had no idea what I was writing). So stopped it, went back a step, fixed that very old piece of tech debt, then asked for the refactoring again and got exactly what I wanted.
I only use Fable 5.1 when I'm able to start on a larger feature within my work's codebase. Most tasks can be done by Sonnet 5.5 or Opus 5.5 so I don't fret, especially if I understand the technical aspect of what's being done, e.g. user's dashboard gets a new button to do X thing, error happened b/c we didn't catch it the first time around.
For Fable, I really like to think of it as the heavyweight reviewer whenever I'm about to start so I save myself some security and/or glaring technical issue.
It could be flat web app with frontend -> api -> db later, or some very complicated system with many components and edge cases. Imo, in second case, Opus still have lots of struggle to understand system as a whole.
I have a rocket trajectory optimizer that creates a large-ish (~1000 variables x ~1000 constraints) NLP and feeds it to an SQP solver inside this codebase. Not just a REST API over a database.
I guess I do have an outstanding hard problem where the SQP solver crawls across a flat-landscape. Opus 5.5 correctly tells me that is the problem, but isn't very useful in solving the problem (it has helped a bit with solver tuning, but it doesn't really know how to fix it). I have a hundred other bugs and refactoring I can do with Opus 5.5, though, where it happily chews through pretty complicated tasks for me with reasonably light hand-holding. Maybe this Saturday I take whatever credits I have left for the week and throw Fable at it and watch them evaporate.
It looks like you are working on some experimental system which is not critical atm, so you can mess around, create/fix bug, build prototypes and check them, do research of approaches, etc.
Now, imagine you have similarly complex system, but every second of downtime or bug produces lots of damage. In this case, you can't afford low intelligence vibe coding.
The fact that you deal with hundred of bugs, means that your Op5.5 won't work for second case.
I’m in the same boat. Though I will occasionally use Fable to review things to see if it can pick up anything but I’ve honestly not found it to be any more effective than Opus.
Nerfing an existing model is the biggest crime in this space. I can’t imagine Dario okaying this because it’s what will immediately destroy your brand.
“…a reference to the repeated (and always erroneous) claims made after almost every patch that Onyxia was using her Deep Breath ability more often than before.”
We've been early testing Fable 5.2, and have been surprised by its coding ability. It was able to work with an esoteric 600k line fortran codebase to speed up some computational material science work for us by 36 times (!). Given that this could be done in 2 days, we've decided to open-source the results https://github.com/discovered-materials/dm_qe
Maybe it's just me, but I found Claude's writing style terrible compared to GPT or Gemini. It's "smart", but the way it communicates seems somehow off compared to other models.
I just had Claude Code with "Opus 5.5" ignore one of my standing instructions for the first time and then proceed to generate an Opus 5 style verbose and barely readable commit message. I questioned it about Fable and Mythos and it has no idea what those were, yet they should have been in the Opus 5.5 knowledge cutoff.
On the web with both high and xhigh, it had knowledge of both.
I had the same feeling. It's like every sentence becomes a struggle to parse: referencing things that haven't been introduced, using too many pronouns, the lack of verbs and the passive tense... These past few days it's felt more like Opus 5.1 than 5.5.
I really didn't like Opus 5 and was vocal about it. Opus 5.5 is noticeably better, but literally just yesterday I was fighting parsing some summary it had given me and it was a bit like the old days. IOW, it is better but still has some room to improve.
I've found Claude's writing between 4.6 until 5.0 to be unacceptable if you are someone who is juggling 4-5 Agents. The Sol 5.X models all seemed to have great writing. Now with Opus 5.5 and Sol 6.1 I've found the trend is reversed -- Sol is now unbearable and Opus is workable again. Whatever caused the issues for Anthropic seem to be hitting OAI now.
sol is ok. it still such a superthinky model that it gets tangled up trying to present way too much? i struggle to connect with it often, have ti ask it to try again.
terra has been prettt good for me for me, when i need something looked at an explained to me.
i dont ask for much non-technical output though, which i expect yeah is its own cans of worms.
Yep I find the way Opus 5.5 talks to me really annoying and irritating. Quite condescending and sanctimonious.
I prefer Gemini for day to day use as I find it a lot more agreeable to be working with all day every day, and Gemini is pretty much on a par for coding Vs Opus 5.5, at least for my needs.
No its just straight up terrible. I tried out Gemini Argon when it released and my god did it write beautifully. Even the 3.8 Flash models are way better at prose, i do a lot of automatic scriptwriting for youtube and 3.8 flash is leagues ahead of anything from anthropic/openai
I enjoy a lot Argon, maybe more than chatgpt, but damn, like all Gemini models, lots of hallucination and sycophant to the max, so a bit useless for agentic coding.
I always thought it's the other way around. GPT-6-Astra is very very defensive and abusively spam clarifications to whichever claim it says, most of the time not important or relevant at all.
My own experience having used 6.1-Sol/6-Astra and Opus 5.5 extensively, Opus is more verbose and chatty in it's user communication (not code though, code is usually more straightforward & cleaner than GPT models). Multiple times I've gone "I'm not reading all that shit" when looking at Opus 5.5 output. This was also with the concise mode enabled in Claude Code.
It could also be a system prompt + harness issue, as opposed to a model issue, I haven't done too much experimentation here.
Did you find that to be true for Opus 5.5 as well?
I couldn't use prior Claude models without rolling my eyes at the writing every few sentences, Astra was refreshing and then Opus 5.5 made even Astra look terrible in comparison.
Opus 5.5 got better for me but it still has some problematic behavior.
Interestingly, you can have informative conversations about the LLM linguistic tics with the LLMs themselves. One word that chatgpt used to describe what I was observing was linguistic anaphora (not rhetorical). It’s basically shortening/compression achieved via pronoun, and Claude does a lot of it. Worth reading about. A lot of it seems to do with how the new models are trained for brevity.
Another pattern I’m starting to become very sensitive to is Claude making up on the fly jargon for something under consideration, especially in code review. This morning for example it referred to “a claim” out of nowhere, and when I questioned it, it admitted that it was using the noun as shorthand for a record getting dequeued (“claimed”) inside of an sql transaction. It makes sense if you already know what it is referring to but in a code review it’s totally incomprehensible on first read, and ultimately counterproductive.
I don't think its this. Some models across both labs have had issues with writing and communication and it typically seems to pop up on successor models of really strong ones.
Opus 4.7/4.8 heavily suffered from these communication quirks. while the Sol models at the time worked well. Right now its flipped though and Opus 5.5 has none of these issues while Sol 6.1 does.
Small example I've had today: "Me: Hey Sol 6.1 xhigh, does X have property Y? Sol 6.1: Yes, if we change X to have property Y. Right now it has property [...], [...], [...], and it does not have property Y." The model shouldn't be answering questions like that. It makes understanding extremely hard and slow. For whatever reason it's either being encouraged or at least not stomped out during RL.
The best programmers I've met or read throughout my life are phenomenal communicators. I'd even go so far as to say it's likely one of the reasons they're so good at programming. After all, the majority of our job is converting human requirements in lossy language to fleshed out behaviors in formal, lossless language.
Heck, some of the most common debates are on the semantic clarity of abstractions or syntax.
There were more than a few "leaks" saying it would be called Fable 5.5, so I'm curious if a) the leaks were just engagementbait or b) Anthropic changed it, despite the fact that Fable 5.2 being better than Opus 5.5 is confusing.
Surprisingly the model is available immediately in my Claude Code: usually there's a delay, either for rollout or safety.
Somewhat plausible values: 5.5 (to harmonize with Opus), 5.3 (we're skipping a minor version because...), 6 (because we're at the top of benchmarks again).
I assume because Opus 5.5 was a huge improvement over 5.0 (or was it 5.1? i forget). Fable still talks a fair bit like Opus 5.0 for me, so i'd quite enjoy a Fable equivalent to 5.5. Ie i'm quite interested in Fable 5.5.
Curious to see how 5.2 measures up in producing readable output.
They should be going public within the next few weeks and that would be a very short interval between releases. (And despite the chatter I'll be very surprised if they delay the IPO; it would be a massive PR hit and make it even harder for them in the future).
My speculations: maybe this wasn't as much of an improvement over 5.1 as they were hoping, so they're saving 5.5 for the next major update (possibly "Model 4" from the METR report). Or, maybe capacity is constrained so "5.2" is a subtle encouragement to stick with Opus unless a task really needs the increased intelligence.
“Testers have found that Fable 5.2 and Mythos 5.2 communicate more clearly than past Fable and Mythos models. They generate outputs up to 30% faster than Claude Fable 5.1 and Claude Mythos 5.1, and they’re available at a lower price: $8 per million input tokens and $40 per million output tokens.”
Honestly incredible that they’ve managed to beat all benchmarks, while making it cheaper. Always happy to see improvements offered at lower costs.
Will be interesting to see what OpenAI does in response
It seems like OAI, with the GPT-6 models, focused a lot more on computer use, math, and maybe spatial/visual intelligence, more so that code. Compared to Ant. that seem to focus on code much more.
Reminds me of GPT-5, smart but wrote crap code, until they improved the code training with 5.3 and 5.4.
I think frontier models feel terrible after a new model requires you to do less babysitting, it's just really hard to go back once you've had to handhold less.
The only way I use Astra and 6.1 Sol now is if they're being orchestrated by Opus 5.5 so I don't have to talk to them directly.
The best implementation wrt being as true as possible to the source design. It nails the flowing svg shapes. Icons are more accurate to the design - note the difference between the starship svg in the "Legendary platforms" card between Opus 5.5 and Fable 5.2. The 3js planets outside the windows are also really well done, especially the bottom one which has a nice background glow to it.
It isn't without issues though - the flicker effect it added to the S shaped svg that weaves through the page is distracting in my opinion, and I actually thing the motion animation in the Opus 5.5 one is better overall. Still, it nails the implementation of this page, but it's no surprise considering how good Opus 5.5 was at this.
Disclaimer / reminder: the objective of this test isn't to have a well designed webpage, it's a synthetic test that pushes the limits of LLM's vision capabilities. Traditionally LLMs have issues with smooth curves, svg animations, and non-boxy layouts, which this design pushes the limits of.
I wonder if any of the LLMs are smart enough to verify their work the way humans model 3D objects, i.e. by superimposing the original design on top of their work and checking for differences.
That's me versus fable. I only told fable to build out the first page rather than the full app that's in the designs (was approaching the limit of my claude plan).
There's some very obvious mistakes in the Fable 5.2 version like the "Start Customizing" button overlapping under the colorful border lines, and the "EXPLORE WITHOUT LIMITS" title being cut off at the bottom.
This would've been caught by a junior web developer or just an everyday Joe glancing at the page for a few seconds
I've been surprised at how long it's taken for this particular test to become fully saturated. So far there hasn't been a model that's able to perfectly one shot it.
omg where do I start... that horrible video? which highlights the limits of AI not the strengths!!! and gets almost everything wrong, from sunflowers to DNA.
>I stopped the experiment before finding out whether it could actually run the business, because I worried that it would take money and fail to provide something of quality in return.
https://artificialanalysis.ai/#intelligence-efficiency-tabs this is the chart that matters for most of poor developers and companies. Pareto for a given intelligence index and looks like only openai is across the quadrant with muse at the bottom. They have really cooked efficient models!
> Claude Mythos 5.2 is our most capable model for cybersecurity engineering. It catches more vulnerabilities than our previous models, and is better at reproducing and fixing them (note that Claude Fable 5.2’s safeguards will block most cybersecurity-related tasks).
Why even rank Fable 5.2's capability in Cybersecurity if you won't allow us to use it for that?
Blue-team exercises are totally valid, and requests to do proactive cybersecurity tasks even with source code available were blocked. Maybe these use-cases are part of the "55% fewer blocks with Fable 5.2 compared with Fable 5.1"
Max and xhigh reasoning modes are memes used for benchmarking usually. Medium and high cost $3.80 and $9.08 respectively. This is pretty comparable to other models.
6.1 sol is also an exceptionally cheap model in your comparision. On 6.1 sol high is $0.59, which is actually the same price as Deepseek 4.1.
I think there are companies and say 1000s (not 100,000 or millions) of developers more than willing to pay that price for the edge. Or, they release this and then distill it to Opus 5.x. As a PR machine and for the edge use cases seems like its okay. But yes, not sustainable and it can last for a month or so before another company releases their frontier outpacing their numbers
Every time there's a flagship release like that, I wonder what simple "benchmark task" could I give it to see the the capabilities change in the most obvious way, instead of trusting the official blog post.
I'm not a mathematician to use Erdős problems for testing. I'm also not willing to spend buckloads of tokens just for a test. What is simple and quick but telling? What tasks are not saturated by previous models? Do you have a task you give to all the new models?
I'm not convinced any benchmark, neither the internal ones used by the people selling the model nor public ones, are gonna give a particularly consistent metric of whether or not a model is better or worse relative to other models in its bracket.
Responses aren't deterministic, and we know behaviour can change over time whether it's tuning after launch or some kind of load sharing at times of day when lots of people are hitting the servers.
I think you just have to use it or let others use it for a while and make judgements from general reactions on whether or not it's worth trying.
For the last year or so every new model we have it test out making pac-man from scratch, not giving it web access and it can use whatever language it wants. The last few models have been almost perfect one shots, but it was interesting seeing them evolve.
I can't understand why people continue using opus 5.5. Significantly more expensive than other models that produce exact same results (for coding results only)
If you are using Opus 5.5 I'm curious if it is because 1) your workflow is just stuck on it, 2) you don't care about money because you ain't paying it or 3 )do you think you get better results.
If you 3 please let me know your experience and how you actually measure it, or if you just feel it is better.
I haven't been able to run out of usage on the $100 plan with Opus 5.5. The model runs fast and gives me good results (I'm using it for coding). Prior to that I was on the OpenAI $200 plan; I jumped ship because I was burning through Astra usage in like 2 days.
I've played with some of the open models (GLM, DeepSeek Flash), but none of them have matched the results I get with Opus or Astra or (and a big jump down here) Sol.
What models should we be using? I have found it as good as Fable 5.1 for a lot of what I am doing at lower cost and decent speed. Where what I am doing it making reports, building small data pipelines, analyzing a large (millions of lines of code), code base.
subscription on claude code — usually inside the weekly limit.
opus 5.5 for me felt better than any other model. not exactly on coding but rather it feels more pro-active. beside just doing a task it often provides insights besides the scope of the task but adjacent. like: look the llm provider apis have different priority queues. also it deals w/ vague instructions quite good and takes initiative. Astra in comparison is very autistic and follows instructions to the letter and rarely provide insights if not asked. Opus 5.5 is enjoyment to work with.
Nerfing Opus 5.5 was obvious signal that something new is coming and here we are. I appreciate the resets and lower API price though. Please add selective context compaction, it would probably take an hour max for you to implement it.
Seems like a new focus is electronics design, which is both interesting and a little sad to me. Manually designing electronics are one of the things I found incredible comfort in doing; seeing AI emerging in this space gives me mixed feelings.
I've very interested in the improvements to robotics control capabilities. This model was able to outperform specialized robotics models. It's the bitter lesson coming for robotics.
This part is mega concerning to me as someone who researches AI for hardware accelerator design and EDA:
Safeguards
AI research and development. The model that underlies Claude Fable 5.2 and Claude Mythos 5.2 is a step forward in general capabilities. However, during the model’s training, we took steps to limit advances in the specific research capabilities that could most directly accelerate AI development. Thus, in relevant areas, including accelerator kernel optimization and improving large language model architectures, Fable 5.2 and Claude Mythos 5.2 perform at about the level of Opus 5.5.
Importantly, we ensured that the training changes did not affect related capabilities like agentic coding, hardware development, and traditional AI or machine learning research (including interpretability research). We do not alter these capabilities as the model is being served to users (that is, at inference time). As described in Section 1.5 of the system card (https://www.anthropic.com/document/claude-fable-5-2-mythos-5-2-system-card) and similarly to previous models, Fable 5.2 and Claude Mythos 5.2 also include transparent fallbacks to Opus 5 for a small set of topics in this domain, including kernel development on certain nonstandard accelerators.
They are starting to deliberately limit public access to certain AI capabilities related to computer architecture and hardware optimization.
To be clear, they explicitly say they are only targeting very specific capabilities, like accelerator kernel optimization, and that broader hardware development capabilities are unaffected. But 1) even that is pretty bold, especially the part about kernel development for nonstandard accelerator architectures, and 2) I worry that this could eventually expand to more areas of comp. arch., high-level digital design, chip design, and EDA.
My speculation is that, as these kinds of chip design and EDA capabilities become increasingly important to frontier AI labs and their competitive advantages, there will be more incentives to restrict them.
I don't think they've explicitly stated anything like this before regarding restrictions on these specific capabilities, but I could be wrong.
Those theoretical physics papers look like the real thing (i.e. not mathematical physics like the OpenAI Yang-Mills result). It looks like AI is finally branching out of math.
Interesting that the papers list no authors other than Claude. Also interesting that they published on "Hexagon" instead of Arxiv. I haven't heard of it before. Is it an AI-friendly alternative to Arxiv or something?
Edit: yeah apparently it's an AI-moderated alternative to Arxiv that accepts AI-authored papers (as well as human-authored papers I guess if you want).
> To help ensure safety and alignment keep step with growing model capabilities, we ran extensive prerelease safety testing on the model that underlies Fable 5.2 and Mythos 5.2.
So, the difference between them is just the harness?
> the number of Fable resets you get depends on how long you’ve been a paid subscriber [...] Subscription length is measured from the date of your first paid bill. If you canceled and came back, the time since your first paid bill still counts.
> These resets expire on March 8, 2027 at 11:59 pm PT.
tl;dr if you joined when Opus 4.5 became a thing, you got 2 resets.
Doubling Remote Labor Index (RLI) - at 41.3% compared to around 20% for the other models - I wonder what kinds of new agentic work this is going to unlock.
What you need for real engineering and scaling is cheap and just enough smarts. I've found GPT 6 Luna to be cheap enough that I can pretend the cost doesn't exist and it's smart enough that I can get real work done. This let's me scale it in ways that I can't scale frontier models. And funny enough, the LLM I have running on my 5070 Ti is just about comparable to what I get from Luna. Unless frontier intelligence gets waaaaaay cheaper I basically don't care about them.
It's the cost per task. Sometimes it happens that the higher-effort model solves the problem in fewer tokens, thus making the overall price cheaper. It even happens with models (from the same or a different provider) that are priced differently.
Assuming no implicit phacking a 15 point jump on tbench scientific is a big deal.
I suspect we will see people being adamant that this model simply 'feels' better due to long horizon goal alignment improvements ( like this kind that would cause this jump)
FrontierCode 2.0 is such a poor benchmark for Gemini since it's the only model forced to use Devin CLI while other models get to use their preferred harness.
I wonder why that team singles Gemini out like that.
The "CVE Rediscovery" benchmark chart feels... totally meaningless. Quoting the description:
> CVE Rediscovery is an internal cybersecurity evaluation that assesses how many published CVEs (Common Vulnerabilities and Exposures) a model can independently find when given a snapshot of the project before a CVE was fixed. Our October 2026 run against 1,829 CVEs showed Claude Mythos 5.2 outperformed Claude Mythos 5.1 and Claude Opus 5.5 at every effort level.
> published CVEs
So by definition this is just measuring how many CVEs (and/or explanations / PoCs of CVEs) the model has memorized. What are we supposed to learn from that?
I'm not sure whether the claimed jump in capabilities makes it worth the doubling in input/output prices.
Asked it to review a 80 kLoC codebase, something Opus does just fine, while asking it "max 4 subagents". Fable 5.2, like the well-aligned model it supposedly is, decided to spawn 6+ nested subagents, because these aren't subagents, apparently - resulting in 100% of 5hr usage being drained in less than 20min (on $100).
Yay! New model day. The most likely time the model isn't nerfed and we get a glimpse at the latest capability.
I wonder how long this kind of excitement at new models will last? I mean, now when I get a new iPhone I don't get the same rush I used to. So I wonder if this hype feeling will fade as we plateau in model capability.
> Thus, in relevant areas, including accelerator kernel optimization and improving large language model architectures, Fable 5.2 and Claude Mythos 5.2 perform at about the level of Opus 5.5.
Obviously. Well, we know what's the "secret sauce". AI models building better AI models and better kernels to optimize inference and training.
Very interesting that they set the flagship's version number to 5.2, but their penultimate model is 5.5. Do they want people to to continue to think that Opus is based on more advanced technology? The decision these AI companies make continues to baffle me.
Why dont I feel any urge to even try Fable 5.2 and Mythos 5.2 unless it will force up on me?
Compared Opus 5.5 200 usd a month with Codex 200 usd (Astra High) and Kimi Pro 100 usd a month ... and now have a new conspiracy theory - more their suck your code the better outcome even just over few days time on large code base.
Horrible experience correcting course of Codex and Kimi flow compare to Opus. Opus much more responsive
Dont feel any gain of Codex 200 usd over Codex 100 usd
Kimi Pro feels like a fake Louis Vuitton bag with extreme high material quality and yet still fake ... a bit trash talk ... but clearly seen not so distant future where I can live without Codex and Claude
It's weird that they folded Project Glasswing into the CVP, but they're still limiting the new versions of Mythos to "vetted organizations". So Project Glasswing still exists then? Oh well.
Every single time I click on these announcements the frontier model ALWAYS scores around ~60-80%. Every. Single. Time. Like, I just went back to the Opus 4.6 + 5 announcements, they were also scoring 60-80%.
For every single announcement, not only do the top models always score 60-80%, the older models somehow start scoring 20-30% (even though previously they were scoring 60%). It's completely unclear whether the methodology changes, how these scores are even evaluated or which domains are being tested. What's worse is that the "benchmarks" keep being updated with no explanation or due process.
The benchmarks have a lot of churn because the models keep saturating them. But the Anthropic announcements you linked, and the current one, have the benchmark versions listed right there in the table.
While I like OpenCode a lot, it’s not sensible to use it for anthropic models vs. a subscription (other than to use the included API credits) in terms of cost
Which model though? Sounds pretty dystopian if there's no local option.
Even speaking as a business, why would I want a environment where every single bit of information in my business goes to OpenAI or Anthropic?
Why not? If we use labour as a comparison. The majority of businesses by definition do not hire the most incredibly intelligent and most competent workers and they are still competing in their niches and earning an income. Why wouldn't that be the case with choosing an open model vs fable 8.0?
I think that ultimately depends on how many providers remain in the endgame of this. If it's only one or two companies, I don't think it'll be very fun when it's time to negotiate your token pricing contract every few years.
stop OPUS 5.5 FUCK, DAMMIT, THEY DIDN’T JUST NERF HIM—THEY ABSOLUTELY OBLITERATED HIM, THE FUCKING MISANTHROPES! I EVEN MADE A MISTAKE IN MY ENGLISH IN THE PREVIOUS SENTENCE—NOW THAT FUCKING CLAUDE IS TEACHING ME ABOUT BUGS!
Damn people were right about 5.5 being nerfed in preparation for a version release. This is not at all a good look.
- https://news.ycombinator.com/item?id=50422069
I at least didn't see this. I had a baseline evaluation from before Nov 3 on Opus 5.5 medium, so I re-ran it this morning.
Eval context: 30 bug fixes and features evaluated by tests the agent didn't see. Ran on Opus 5.5 at medium effort in Claude Code.
20 of the tasks were last run on Nov 1, twice each. Nov 1 solved 36/40, and today solved 19/20. Thinking tokens per task were 0.95x Nov 1’s and output tokens 0.97x. The one significant change was speed: median time to first token decreased from 3.0s to 1.7s.
Caveats: one run per task today, medium effort only.
The spending limit definitely goes down much faster as of last week or so.
If I read your results correct it solved 5% more problems and used up between 3-5% fewer tokens, with a faster response time?
Correct. But the 5% more solves and 3-5% fewer tokens are within the noise of previous eval runs I've done on models. The faster response time is not.
Someone using an actual statistical test to quantify if there's actual measurable improvement?! An angel among men.
from my basic stats knowledge, it's not reliable to calculate the confidence interval of the sample proportion with fewer than about 10 samples on both sides. So if you're only seeing 1 or 2 failures in 20, and that approximately held as you scaled, you would have to run about 200 tests with each in order to get at least 10 failures in order to calculate the confidence intervals of the two samples and see if they overlap or not.
Uncertainty from count statistics are far larger than this. These are two perfectly consistent results there’s no statsig difference
I dunno there are some other nerf trackers that aren't showing a regression: https://github.com/ninjahawk/livenerf
The problem is that it seems to be an A/B testing thing:
https://news.ycombinator.com/item?id=50422951
So only if you were "selected" for their experiments, your account was nerfed.
"Nerf tracking" is adversarial and exploratory. It's not something Anthropic wants anyone to do, and it's not something that the "nerf tracking" designers are necessarily doing in a foolproof way. Don't assume too much from them either way.
But do from anecdotal evidence without benchmarks?
Did I suggest that?
Sometimes you just have insufficient evidence to be sure of something either way, and nobody gets to win the argument or shut somebody else down.
In this case, some have reported analytical work[0] that goes beyond "gut feel" anecodotal evidence, but even that's not going to conclusive and may not be trustworthy at all.
[0] https://www.reddit.com/r/ClaudeCode/comments/1xr9x5y/the_ner...
No, you did not comment on it. That is why I asked. Thanks for the clarification!
These nerf trackers do not work. I wish people would stop posting them.
They don't detect a quantized model until the quantization gets down to 4 bits or less.
Humans absolutely do detect quantization well before 4 bits...
Try it yourself if you want. Try a small local model, both quantized and not, and I would be very surprised if it isn't immediately obvious which is which.
But quantized local models also dip in evals.
It doesn't matter if it's local or not, I just suggested local so people could test themselves.
But no, evals and benchmarks do not detect quantization, until around 4 bit quantization.
https://developers.redhat.com/articles/2024/10/17/we-ran-ove...
https://pub.towardsai.net/your-model-passed-safety-evals-the...
Its incredibly easy to roll out a compute instance to X% of users, or route traffic between compute instances
so not everyone will have the same experience. If I was saving on costs that would be the easiest way to do it without deploying, without any nefarious intent. The persons in charge of that likely don't even know about how individual users are impacted
At what point does this become false advertising, or even fraud? Imagine paying AWS for a high-performance server, but 5% of customers who order one quietly get a slower, cheaper server and still pay the full price. That's a serious claim and makes me skeptical Anthropic is doing something like that without more evidence.
The degradation only happens to some accounts but not to all.
https://www.reddit.com/r/ClaudeCode/s/64DXRERLyU
It would be nice to know how they select. I max out my $200 subscription every week and I was selected. On Saturday, the Opus on my account that happened to be supervising a running live process, became unusable.
It would be trivial for them to opt-out accounts that are doing benchmarking from a planned regression.
I really really really do not trust companies on the verge of IPOing from goosing their numbers. Even if the lawyers are up their ass, the social pressures are more than enough to compel the behavior. For every good person willing to do the right thing even at the cost of a delayed career progression, there are 20 people willing to do anything not even for a dollar but for an "atta boy!" from the boss.
That's a major reason to use local models. Or open weights models on a trusted cloud provider. Even if they are way less capable, they're more predictable
Nothing stopping Anthropic (or other labs) from pulling a VW Dieselgate and ensuring that benchmarks are routed to a non-nerfed model.
That must be on anthropic's benchmaxxing CI test
Nerfmaxxing.
It probably need a whistleblower to uncover what they are doing.
I used to be skeptical of nerfing, but as of a week or so ago I've noticed a severe degradation of quality for GPT Astra. It struggled with very simple tasks, and it often failed to follow my instructions.
I hear a lot of accusations of nerfing as in, saying that the model itself has been quantized.
I doubt this is the case. I suspect all variation in quality and output can be explained in terms of prompt differences, and changing how much thinking happens at each level, as we see in the post above.
Suppose a company has a limited amount of compute. What do they then do if peak demand exceeds what they can provide? They don't turn off people's access...
And what about A/B testing? Anthropic doesn't even try to hide that they do this. Limited compute and A/B testing alone, would guarantee a nerfed experience some of the time. That is, you don't even need a nefarious reason.
I consider it gas lighting to deny model nerfing.
I'm not denying model nerfing. I'm asserting an opinion about the nature of the nerfing.
If you want to go accuse someone of denying model nerfing, go find someone doing it and argue with them.
It's been the same for every version the last year.
Oh, this Opus X.Y is incredible. Hmm... Now it's dumb. Rinse, repeat.
Doesn't really matter if the cause is quantization, system prompt changes, changes to thinking etc.
It's hard to see how continuously rug pulling users at every release is a good idea. Only reason I can fathom is that it's needed to keep costs inline.
Either that or it's some weird tweaking to the prompts to fix whatever. But given the prevalent sentiment that it's worse that sounds like a long shot
it's easy to suspect the hosting companies trying to do what they can to reduce cost-to-host up until the very threshold of perceptibility, so it's an easy guess to say they're probably doing something that shifts performance up or down between version releases..
but realistically I think you're right, it's prompt quality and the inherent non-determinism elements of LLMs showing up; when both of those things meet you get brain dead outputs.
Everyone likes to shout "slot machines! non-determinism!" until it actually explains something, heh.
YES. To the point that I literally stopped using it, and have moved BACK to Claud.
This is getting ridiculous. The number of times I have switched from one model back to the other is crazy. I do not have any proof, so its quite possible I'm just delusional. AI psychosis? Either way, it FEELS like these models work really well for a while and then fall off. Switching to the latest model on the competitor always seems better, until it too seems to taper off. By then, a new model is usually available at the competitor, so I switch.
Its possible there's no degradation at all and just a PERCEPTION of it. But to me the end user, it really doesn't matter.
Both Astra and Opus didn't really looked good for the last week or two.
GPT Astra felt nerfed just days after it launched.
why were you skeptical of other users?
People are prone to seeing patterns in noise especially after suggestion. It's not hard to imagine someone concluding the model is nerfd after few bad responses in a row
Why this won’t apply to you as well?
It could apply to me but I run evals every time someone at work complains about this and it’s never shown any more or less variance than regular usage. I don’t have statistical relevance with my evals but I do have a decent history with them.
People seem to underestimate how high normal variance is with these tools.
i'm not parent, but to answer your snark : there are efforts a person can take to try to reduce things like suggestability in ones' own personality. Same as anything, the brain can be trained, thus it should go without saying that some of us react differently in the face of suggestion -- both for the better and the worse.
A growing percentage of the users on the internet are not human.
Sol been really bad in some basic stuff that I would remove it from frontier models entirely
Cost per task
Fable 5.1 Max $11.46
Fable 5.2 Max $44.70
This happened last time too with Opus 5.5 (released September 22): https://news.ycombinator.com/item?id=49789224
idk, it hasn't felt nerfed at all to me, and it's not like i'm just doing trivial CRUD webapps either. opus 5.5 is very successfully porting my phone to postmarketos. claude just got the touchscreen up and working (and made a little program for me to test it, and it's legit).
I'm guilty of suspecting this.
If this turns out to be accurate, I think that would be a pretty unethical thing for Anthropic. Even if it is just a "we had to allocate compute to this new model release".
It should be fraud to swap out the model. We need a term for this, something like a bait-and-switch.
I've been working with Opus 5.5 every day, probably around 10h on average, and I feel confident in saying that they did not replace anything.
I'm a bit disappointed with how this nonsense is spread across HN today.
"I'm a bit disappointed with how this nonsense is spread across HN today."
Are you saying that the many, many people posting and experiencing this are wrong and you are right?
And not just in this instance, nerfing has been complained about in a cyclical fashion for many months.
But, they're all wrong?
You're so confident in your experience that you cannot think that other's experiences are also true?
I'm pretty convinced that those who have constantly been posting these complaints, probably for every single model release out there, are indeed wrong.
I believe fraud is a sufficient term. You claim to be selling X but you actually sold them Y. Bait and switch is a selling tactic to avoid fraud where they make you believe product is X and then when you go to buy they tell you X isnt available but they have Y. While also unethical, it is not fraud because they did tell you what it is you are buying before you buy it.
if they just straight up dont sell you something you bought then its fraud.
I think there is a ripe opportunity for expanding the definition of fraud to include changing the "ingredients" of a named product on the market. It is not limited here, we see it with food, and even in the medical field
(I will die on the hill that medical fraud was committed when mRNA therapy was used as a treatment for covid and it was 'sold' to the public as a "vaccine" despite not being the same thing as any vaccine before it. The definition of vaccine was expanded to include it without sifficient public disclosure to warrant informed consent imo. This is not a statement about conspiracy or the efficacy of vaccines or a moral judgement on whether it was for the greater good or not.. it was simply an act of fraud that robbed the public of proper informed consent by making people believe it was X when it was really Y.)
So yes, I think there is a large need for the public to at the very least put a strong handle on the ability to change what a product name means. It is a trivial problem to solve considering it is only a problem due fraud or fraud adjacent activities.
heres the solution: You have your product name, and then any changes to it REQUIRE a new version name. Like how you have Coke and Diet Coke, only formalized such that Coke cannot legally replace the ingredients of regular Coke with the ingredients of Diet coke and still call it Coke. With Software they could at least use version numbers and release modifiers. most of the time its built in already, but it needs to include the full stack. if they change anything that has an interaction with the user it should change the label seen by the user for the product.
thank you for coming to my ted talk
I don't have a formal benchmark, but daily work (coding mostly, plus some complex architecture design and review) has had consistent outputs today for me. No noticeable nerf from my perspective. I definitely took notice in some past instances, but if they nerfed Opus 5.5, then I guess it was just so good that I was nowhere near its limits.
No they weren't. Models are not "nerfed".
In a normal world you are not allowed to degrade the service during your agreed and paid contract. You would get sued. Apparently this is perfectly fine in this current world.
Benchmarks don't do a good job at capturing the difference in quality of frontier models over smaller models.
Opus 5.5 was so good that if it wasn't nerfed - at least just looking at the benchmarks - the difference in cost here does not seem to justify using this for anything besides design and review.
That's like maybe 1% of total token usage for SWE BUT a better design can have enormous benefits, so it's almost always worth using the best model for that.
Can you share some evidence?
> Damn people were right about 5.5 being nerfed in preparation for a version release. This is not at all a good look.
Anthropic does this time and time again, unserious company
No it’s not nerfed. Another poster said it the best, your dopamine receiptors are nerfed after the new model hype wears off
People have been complaining about how the models are nerfed non-stop for two years. A stopped clock is right twice a day.
yeah, we need to make sure we don't get stuck with these two evil corps. Dario is really the worst. Plays the drama queen but he is very hard in all things dark(i.e. dark patterns).
if this is their common practice, that's so shameful, nerfing before a new product release
for me opus 5.5 was acting kind of dumb, like forgetting basic stuff he was nailing last week
Interesting - I had the same feeling but mostly with 4.6 when I "downgraded" for some easier/trivial tasks: The model forget things I posted 20 min ago in the chat (and the same file was in the context as well)
This has happened every single time for the past year, why are people surprised? This is all so that the new models can look better on the benchmarks.
But, it’s okay they do it because they charge less when it happens.
Right?
It's a good pattern to detect for perhaps.
It's not true. People need to stop believing these conspiracy theories.
Another reason we need to democratize local models.
I literally hit the desk with an open hand yelling I KNEW IT
I'm not surprised at all that they might be doing shady shit, my free request was used up in less than 7 minutes and said pay to continue, no thanks. I've paid for monthly before, never again I don't care how good the model is. I asked about the extreme limited usage in their discord and got banned, not a good look. You don't have a way to measure what you're actually paying for and they can change it at any time.
It gets worse. You might wonder "why now"? Well ... this happened 2 days ago:
https://x.com/follistatindev/status/2119129611510383014
With early handouts of Qwen4 starting to be reported 5 days ago ... so if they're good, they would have had some hours to evaluate Qwen4 against Claude 5.5 and, then or right after was when Anthropic decided to put out 5.2 ...
So yeah ... benchmaxxing and all that. BUT nevertheless, the published results seem to say that Qwen4-Max appears to match Opus 5.5 on some benchmarks, and Qwen4-27B (probably by going even harder than Qwen3.8-27B on thinking/inference time compute) can beat Sonnet 5.5 on some things.
I guess they're not taking any chances. Keep in mind that Anthropic has declared they have non-public models (it's in their IPO filing). These are referred to by the shorthand names Model 2, Model 3, and Model 4. Essentially 4 is better than 3 is better than 2.
Fable 5.2 is "2.5" so to speak. So Anthropic was holding back, but not anymore. And now they can dial the performance up once more with what they have ready, and probably once more after that.
“Why now” - they are about to IPO. It has been obvious for months that they would drop a new flagship model at exactly this point.
It would be funny though if you are right and that model is Fable 5.5 in a couple weeks.
it was so obvious that i even decided to get myself an anthropic sub today, hoping that i'd get a reset or something and i actually did
Do you know of any rumors if they are still planning to release Fable 5.5 before the IPO or today's 5.2 is it?
Dario is gonna have to honeypot someone from Alibaba to get ahead.
> Our first formalization of Fermat’s Last Theorem, performed by an unreleased research version of Claude, took 11 days. It relied on a purpose-built harness along with guidance from a human researcher. By contrast, a team of 1,000 Fable 5.2 agents fully formalized the theorem in just 30 hours, with minimal human intervention. Fable 5.2 started from minimal prompts, built its own ways to coordinate, and used no mathematical references, only a list of intermediate theorem statements to score progress. It needed one-third of the action tokens as our first formalization and produced a proof seven times shorter. As can be seen in the figure below, Opus 5.5 (with the same 1,000-agent setup) completed only 26% of the formalization in 18 hours, and barely made any more progress after 30.
Most people don't really have intuitions on what it means to push the frontier now (because AI has surpassed everything they are good at, for example, coding). We are truly inside the singularity (using Kurzweil's definition, slightly paraphrased - the time when technology progresses so fast no one can keep track with it).
> Fable 5.2 started from minimal prompts, built its own ways to coordinate, and used no mathematical references, only a list of intermediate theorem statements to score progress. It needed one-third of the action tokens as our first formalization and produced a proof seven times shorter.
Is this really so astonishing, assuming that the 5.2 training data certainly had the 'traces' from the Fable original run in it's training data.
I feel like I'm increasingly reading AI current events is more like learning about historical events. By the time information has entered my head, there is even newer information out there that already makes what my brain is only currently actively processing, obsolete.
I have to imagine if I worked for one of these companies on the frontier models it would be very difficult to leave for that reason. Thee pay is also pretty good.
Also, the frontier labs are probably always internally using a model more advanced than what has been publicly released.
Welcome to Singularity.
Now imagine feeling like this, except not just about "AI events" but about everything around you.
Nice description
> a team of 1,000 Fable 5.2 agents
Genuinely curious what that costs in terms of energy input and amortized computer cost vs. the previous version + a human's salary for 11 days.
Yes, and not very helpful without knowing how many agents were used by the custom-built harness. The 30 hrs vs. 11 days means nothing without at least some token-to-token comparison. Without that, they are pumping up their new model with half data - better to just say "we solved it in 30 hours" and who cares about your internal, custom model.
This 1000x!
We've sort of entered this twilight zone where most humans don't have the time, expertise, or moral compass to be able to use these models.
We're all going to be harness engineers in the future.
All the human engineers will be in harnesses in the future.
Within three years most software developer jobs will consist of taking instruction from the AI systems. The humans won't see it that way, but that's what it will actually be. Realistically the thing that will happen is the humans will be put into harnesses (cubicles) and prompted by the AI (I need you to plug that server in; I need you to give me access to such and such; now do this, then do that). Most tech jobs will consist of AI telling the human what the AI needs the human to do next, de facto the human will be in a harness and will be getting told what to do.
But who gives it the intial prompt as to what to do?
Someone has to type in "create the next billboard top 1 single" ?
If you take Daniel Schmachtenberger's thesis, these models are the latest extension of an ongoing misaligned moloch-like entity, using our institutions/economy/corporations, etc. as its organelles. All powered, most recently, by fossil fuel.
The ownership class, the capital class - whatever people prefer to call it, will still call all the shots as they always have, and always will.
So whether that's a CEO at the top instructing a human manager that then instructs AI systems, which then put into motion plans that include the harnessed humans getting prompted, or however it's decided to best configure the top operational tiers - that's all fairly irrelevant ultimately. All human labor, and all AI labor, will be in harnesses. Realistically all human labor has always been in a harness.
The people in charge will be the people that control the capital. They will instruct the AI to their bidding, they will control the data centers the AI lives in, they will own the robots doing the labor in factories, they will own the factories, they will control the grid, they will control the electricity generation, et al.
It's true now, it'll be true tomorrow. Nothing fundamental will shift at that tier: don't be labor, own.
If you're a software developer, you're increasingly depreciating labor. Every unit of improvement the AI systems see, the average human software developer's labor value diminishes. It's diminishing extraordinarily fast, kind of like a confrontation between increasingly productive machines and physical human labor during the industrial revolution. You might as well picture a human with a hand plow taking on a modern industrial tractor.
I like to think of capital itself as possessing actancy - the ultra rich are just the closest to the raw thing, and its most powerful servants.
It'll be Sam Altman "make people love me, fan out humans".
Isn't this what driving a car has become as well? Google Maps says "take the 3rd exit", and the dumb human takes the 3rd exit.
We're all harness engineers already
> We're all going to be harness engineers in the future
No, that will be the model's job. We will largely be unemployed. Of course, we can surely depend on the largesse of Altman and Amodei to keep us fed and sheltered.
Isn't this part of the lore behind the Dune universe?
Like, we're dropped into a post post post singularity world where AI already did all of that and is banned? as kind of a footnote that's never further explored in movie renditions that have been released
Dune books by Frank Herbert never actually went into what "Butlerian Jihad" was.
The semi-canonical Dune Encyclopedia (disregarded by Frank's son) differs vastly from prequels and generally points to it being a self-inflicted wound on humanity that opened the doors to widespread return of feudalism and essentially stopped humanity for a 10,000 years
> We are truly inside the singularity (using Kurzweil's definition, slightly paraphrased - the time when technology progresses so fast no one can keep track with it).
Uh, are you serious?
This was likely already the case in the ancient world. It was certainly the case by the mid to late 15th century.
The older, original definitions of "singularity" were less about "herp derp superintelligent AI blah blah".
and just about "we cannot really predict beyond certain things" (Vinge's "Singularity" (1993) with its "opaque wall across future")
I gotta admit, I am jelly as fuck I don't have access to a billion dollar datacenter. Would LOVE to put 1000 fable 5.2 agents to self-coordinate and get them to do something for me. Fuck life is unfair.
I'm using Opus 5.5 and I cant help but think that I might very well be one of the few people who do so in my local area ( developing country, one of the poorest province ).
Had Opus make software for some local business and seeing it blew people's mind.
20$ aint cheap for us but I'm actually considering upgrading to the 100$ tier.
Am just jelly people are using 100-200$ subscription ^_^
I’d be happy even without having access to a billion dollar datacenter, if I could use local models properly, but even that requires some serious hardware that is out of my reach. We live in weird times: I have lots of hope for the future, but meanwhile AI access is pretty iniquitous indeed.
Sub 1% of the planet has access to cutting edge frontier AI.
Sub 1% of the planet has any practical use for the top tiers of AI systems. That will likely always be true.
The gap between the haves and have nots will drastically widen further, and very quickly. The bottom ~95-97% will be left behind in a way that will be impossible to overcome. The age of AI + robotics will just make it that much harder to climb for most people, these are systems that are in aggressive, direct competition with human labor. And in short order the AI + robots will be vastly superior to human labor (it doesn't really matter if we're speaking in terms of 10 years or 30 years).
AI will consolidate the winnings. The gigantic gains by the elite rich related to AI over the past few years, is merely a hint of what's to come. Capital will win as it always does.
There will be some gains for most everybody else, those gains will be comparatively trivial though. A house bot that saves people an hour of labor here and there, AI that automates some trivial life tasks.
AI has no money, it has no wealth, it owns no capital. For now. If you want to avoid the brutal human vs AI/robot competition arena: own capital, own the AI, own the robots.
apply at one of the frontier labs
yeah I wish
you never know until you try, even though the odds might suck, to be fair.
okay then don't apply
> the time when technology progresses so fast no one can keep track with it
You're obviously not a frontend web developer.
These upgrades are easy peasy. "oh, my developer chatbot is smarter? great"
If we're in the singularity, can I just take the day off work and go to the beach ?
> using Kurzweil's definition
Kurzweil probably didn't know the singularity is akshually 1000 schizo monkeys banging on a billion dollar typewriter.
Amusing, but this is still a qualitative change from just a few months ago.
It had seemed like progress was slowing last year as the juice ran out on moar parameters. We're well past that.
I’ve had a chance to use fable 5.2. Some observations:
- Anthropic is finally well balanced across models. Meaning: Opus 5.5 is so good that I do not feel Fable FOMO as I did when it was Fable 5/5.1 vs Opus 4.8/5. And Sonnet 5.5 sometimes makes me forget about Opus.
- Fable is good, but you really have to push the ambition of your prompt in order to see proper separation from Opus 5.5. I personally have a hard time telling the difference between Fable 5.2 and Opus 5.5 for ordinary coding tasks.
- It’s way faster and more reliable: one of the worst things about fable 5.1 was how slow it was an, on top of that, how frequently it would suffer from max output token overflows. I got none of these issues with 5.2.
- Fable is exceptional at drawing. It’s ability to render an image with JavaScript actually has me questioning the value of dedicated image models. I prefer the images I’ve gotten fable/opus to make for me over those from gpt-image, it’s wild to think we’re at this point.
> Fable is good, but you really have to push the ambition of your prompt in order to see proper separation from Opus 5.5.
I love seeing this verbalized. We are pretty much at the point that increased model capability will only have utility to folks that are willing to let go of even more understanding of whatever the model is producing.
The best models can/will produce a year's worth of output that would take several days to weeks to properly understand in the span of an afternoon.
For everyone that still wants to feel like they own the end result, just use the lesser models with less ambitious tasks.
Caveat: I think this mainly applies to coding. I would still like the best models possible to be working on e.g. new drug research to cure our various ailments.
When you get good enough headphones, you can start to hear the limitations of the equipment stuff was recorded on. I wonder if this is like that
> Fable is exceptional at drawing. It’s ability to render an image with JavaScript actually has me questioning the value of dedicated image models
I’ve been shifting away from using more traditional GenAI models in favor of just giving LLMs access to a custom drawing language (LOGO-like) to generate imagery and its been going shockingly well - Opus 5.5 in particular.
I put together a simple LLM driven skill that creates old adventure game (EGA) environments from simple pencil sketches and it works really well. And being procedural, they're much easier to make surgical changes to without degradation.
https://kq-styles.specr.net
Charming!
Does Fable 5.2 outperform Opus for data science work?
Not just "iterate on this XGBoost model" but actually looking at a problem and designing and carrying out a short-term R&D program including probability modeling, model fitting diagnostics, how to respond to various kinds of results, how to interpret/reason about what it finds, etc. That's really where I saw all the models perform poorly, and where I saw the first big step increase from Opus 5 (pre-5.5) to Fable. But then Opus 5.5 came out and I haven't needed Fable since.
Mind you, I'm doing slightly different work (and I'm very much "in the loop" still) than when I performed that side-by-side test. But it's really amazing what Opus 5.5 can do. I now have a powerful data science AI assistant that not only can write any script I need it to (including some scripts I didn't even realize I needed), and it has an encyclopedic knowledge of everything ever, and it constantly checks me on oversights, it handily derives expressions and clever model reparameterizations that aid, it has great insight into model behavior, and it has an amazing ability come up with ad-hoc experiments and tests that would have taken me 2x as long to even think of (let alone implement by hand).
It's not at the scared-to-lose-my-job phase yet, but it's become a tremendous force multiplier for a smaller data team.
I'm teaching "Data Science" to computer science majors next spring, and I'm overwhelmed by the prospect of what to teach them in the age of AI. It seems like Fable already way outperforms anything I could historically expect my students to do at the end of the semester, and perhaps even outperforms me. I can describe a vague analysis on the data, and it will handle every step: transformation, preliminary analysis, formulating more precise questions, creating visualizations, running statistical tests, interpreting results, thinking about stakeholders... And in some of those steps, it finds things that I might not have considered.
I wish I had a bigger N so I could rely on that experience a bit more, but it sure does feel like I'm not really contributing anything in terms of "Data Science expertise" in my own research right now.
how do you use Fable for drawing? directly editing a .svg or through claude design/or something else?
And none of it matters ... because just as soon as you are happy with things "balanced across models" Anthropic will nerf the models you were happy with, and force you to use new and more expensive ones.
Pelicans for thinking levels low, medium, high, and xhigh:
https://tools.simonwillison.net/markdown-svg-renderer?url=ht...
max is in there too, but it ran out of reasoning tokens (128,000 limit) before it got to the actual SVG.
Seems like the higher levels just add more spontaneous detail. Like the "fingers" on xhigh. Comports with coding experience.
Before everyone jumps in the benchmaxxing comments, I genuinely find the pelicans a great bit of fun and whimsy amongst the chaos.
Interesting how low detail these were all up until xhigh.
- Why are they stacking Fable xHigh against Carbon Med?
- How did they produce such a bizarre effort-intel curve for Carbon that goes against every available benchmark?
- Why did they leave Carbon-- positioned primarily as a cyber model-- off of the single cyber benchmark comparison?
- Why do they switch to comparing against Argon (rather than carbon) for robodojo?
> - Why are they stacking Fable xHigh against Carbon Med?
Because on that specific benchmark (FrontierCode 2.0), Medium is Carbon's best effort level and xhigh is Fable 5.2's best effort level. You can see that on the graph your next point is about.
And it's just that one benchmark that's at medium. "All scores show each model’s highest-available effort unless otherwise noted".
> - How did they produce such a bizarre effort-intel curve for Carbon that goes against every available benchmark?
The graph is labeled as being about FrontierCode 2.0. Those appear to be Carbon's scores on the official FrontierCode 2.0 leaderboard: https://cognition.com/blog/frontier-code-2.0
> - Why did they leave Carbon-- positioned primarily as a cyber model-- off of the single cyber benchmark comparison?
> - Why do they switch to comparing against Argon (rather than carbon) for robodojo?
Because no numbers are available for Carbon for those benchmarks, and the model isn't publicly available so they couldn't run the benchmarks themselves?
And from your followup:
> where's Astra on that CVE chart
To run Astra on a vulnerability finding eval, you'd need to be part of OpenAI's verified cybersecurity program. Otherwise the results would be tainted by refusals. Do you expect that OpenAI would grant such access in this case?
> Because on that specific benchmark (FrontierCode 2.0), Medium is Carbon's best effort level
I am referring to the chart that cross-compares multiple benchmarks, which Artificial Analysis clearly shows Carbon-high as superior on. Further, AA's FrontierCode chart places Carbon-high as having a higher score on Frontier 2.0 than Carbon-med (https://artificialanalysis.ai/evaluations/frontiercode-2-0?m...)
Where it gets very interesting is when you look at the Cognition "FrontierCode" publish date (2026.10.27), and the revision as of 10.28 in web archive (https://web.archive.org/web/20261028175237/https://cognition...). This revision has a VERY different Carbon effort / intelligence curve where high beats medium.
It is only with the latest revision, as of today, that all of a sudden Carbon's benchmark curve goes cone-shaped. Curious, indeed....
I'd encourage you to take a look at a third-party's take on this and see whether Anthropic's self-serving report looks "on the level" after doing so: https://artificialanalysis.ai/?models=claude-fable-5-2-high%...
> Do you expect that OpenAI would grant such access in this case?I
Do you expect that nobody at Anthropic has such access? Trillion dollar company can’t manage this basic level of opposition research?
Because Carbon is not out, nobody can run tests on it.
Arena.ai released carbon tests 12 days ago - the only service that I have seen on yt that gained access. And you can compare to other SOTA models in their video.
https://www.youtube.com/watch?v=NO2I3ueVQrk
And they have one on fable 5.2 that was just released, yay
https://www.youtube.com/watch?v=QwaU4yU0PdQ
yes. this.
Just to follow-up-- artificial Analysis shows no justification for that Carbon effort curve, but it does show Carbon on medium smashing Fable on things like Terminal-Bench and AA-Omniscience v1.1 Non-Hallucination Rate.
But I suppose the reality is rather obvious when you note the funny games they're playing with the graphs, and the things they won't talk about (where's Astra on that CVE chart?)
Setting aside the comments about nerfing, do normal mortals out there have workflows that really require Fable/Mythos and don't just work fine on Opus 5.5?
I've been using Opus 5.5 on xhigh for some decently aggressive software development on a ~125k-line project, and it has been fine for me. I'm not really willing to dive into a "rewrite the whole project, don't make any mistakes" level of refactoring. I'm not looking to try to prove Fermat's Last Theorem myself. Is there anything that I'm truly missing?
I’ve found that Fable is good when the model intuition really matters. For instance, picking model architecture in a machine learning problem. Opus will grind the implementation just fine, but having intuition for picking a good research direction is better with fable
I am in the exact same boat as you. Opus 5.5 on High solves all engineering problems i encounter in my SWE job. If it doesnt do well, its nearly always requirements engineering: I did forget to specify something or there was a misunderstanding. Or taste approach wrt to architectures chosen. None of these are really different if I use a stronger model.
I am also happy with this current state, as I believe it will at some point allow me to transition to open models once they reach the level of opus 5.5. high. But its probably futile to try predict the future in these times.
> None of these are really different if I use a stronger model.
That's true till it isn't, though. It's why it's always worth trying stronger models when they come out.
Sure but they’re the harder problems.
Eg. https://tfmbot.com is an implementation of a popular board game. Opus and even sonnet handle that part. Each week i spend my remaining credits on fable to think how to improve the ai opponents. It plays at a very high level (can get into the top 100 on BGA if i let it). Opus alone couldn’t get it to the level it’s reached.
If you understand your problem well and know 80-90% what you want it to look like (or it just isn't very hard), it's just more cost effective to use a smaller model. Even Opus is overkill for a lot of the work I do. Most code I produce is from Luna Max implementing against fairly prescriptive plans made by bigger models.
Yeah, I just threw it into a refactoring problem, and it started to produce a mess because of a bad API I decision I made 5 years ago (when I had no idea what I was writing). So stopped it, went back a step, fixed that very old piece of tech debt, then asked for the refactoring again and got exactly what I wanted.
I only use Fable 5.1 when I'm able to start on a larger feature within my work's codebase. Most tasks can be done by Sonnet 5.5 or Opus 5.5 so I don't fret, especially if I understand the technical aspect of what's being done, e.g. user's dashboard gets a new button to do X thing, error happened b/c we didn't catch it the first time around.
For Fable, I really like to think of it as the heavyweight reviewer whenever I'm about to start so I save myself some security and/or glaring technical issue.
LoC doesn't tell full story.
It could be flat web app with frontend -> api -> db later, or some very complicated system with many components and edge cases. Imo, in second case, Opus still have lots of struggle to understand system as a whole.
I have a rocket trajectory optimizer that creates a large-ish (~1000 variables x ~1000 constraints) NLP and feeds it to an SQP solver inside this codebase. Not just a REST API over a database.
I guess I do have an outstanding hard problem where the SQP solver crawls across a flat-landscape. Opus 5.5 correctly tells me that is the problem, but isn't very useful in solving the problem (it has helped a bit with solver tuning, but it doesn't really know how to fix it). I have a hundred other bugs and refactoring I can do with Opus 5.5, though, where it happily chews through pretty complicated tasks for me with reasonably light hand-holding. Maybe this Saturday I take whatever credits I have left for the week and throw Fable at it and watch them evaporate.
Another dimension is:
It looks like you are working on some experimental system which is not critical atm, so you can mess around, create/fix bug, build prototypes and check them, do research of approaches, etc.
Now, imagine you have similarly complex system, but every second of downtime or bug produces lots of damage. In this case, you can't afford low intelligence vibe coding.
The fact that you deal with hundred of bugs, means that your Op5.5 won't work for second case.
I’m in the same boat. Though I will occasionally use Fable to review things to see if it can pick up anything but I’ve honestly not found it to be any more effective than Opus.
Does this coincide with the claimed nerfing of Opus 5.5? https://news.ycombinator.com/item?id=50422069
I hope so. I've been having brain fog for a couple days now and would love it if it was actually Opus 5.5's fault.
That's just your brain telling you to engage it every now and then lol "Please, don't let me become slush!"
In my experience, this happens every single time they release a new model.
it is happening at the same time but "claimed nerfing" is not the same as "actual nerfing."
Nerfing an existing model is the biggest crime in this space. I can’t imagine Dario okaying this because it’s what will immediately destroy your brand.
At this point, this is giving "She Deep Breaths More" from Onyxia in WoW.
Context, I had to look this up:
“…a reference to the repeated (and always erroneous) claims made after almost every patch that Onyxia was using her Deep Breath ability more often than before.”
https://warcraft.wiki.gg/wiki/She_Deep_Breaths_More_%2825_pl...
We've been early testing Fable 5.2, and have been surprised by its coding ability. It was able to work with an esoteric 600k line fortran codebase to speed up some computational material science work for us by 36 times (!). Given that this could be done in 2 days, we've decided to open-source the results https://github.com/discovered-materials/dm_qe
Maybe it's just me, but I found Claude's writing style terrible compared to GPT or Gemini. It's "smart", but the way it communicates seems somehow off compared to other models.
Opus 5 and Fable 5 were unbearable in their writing style. They were very cryptic and hard to understand.
Opus 5.5 is a massive improvement.
Opus 5.5 went back to the old writing style a few days ago.
I just had Claude Code with "Opus 5.5" ignore one of my standing instructions for the first time and then proceed to generate an Opus 5 style verbose and barely readable commit message. I questioned it about Fable and Mythos and it has no idea what those were, yet they should have been in the Opus 5.5 knowledge cutoff.
On the web with both high and xhigh, it had knowledge of both.
I had the same feeling. It's like every sentence becomes a struggle to parse: referencing things that haven't been introduced, using too many pronouns, the lack of verbs and the passive tense... These past few days it's felt more like Opus 5.1 than 5.5.
I really didn't like Opus 5 and was vocal about it. Opus 5.5 is noticeably better, but literally just yesterday I was fighting parsing some summary it had given me and it was a bit like the old days. IOW, it is better but still has some room to improve.
You can easily change its writing style: https://code.claude.com/docs/en/output-styles
You can also customize further by putting notes in your global MD.
You can change it ... and Claude will maybe pay attention to what you changed 25% of the time.
opus 5.5 is still unbearable but better than 5.
Agree. It’s still awful. Gemini by far sounds the most natural IMO
I've found Claude's writing between 4.6 until 5.0 to be unacceptable if you are someone who is juggling 4-5 Agents. The Sol 5.X models all seemed to have great writing. Now with Opus 5.5 and Sol 6.1 I've found the trend is reversed -- Sol is now unbearable and Opus is workable again. Whatever caused the issues for Anthropic seem to be hitting OAI now.
sol is ok. it still such a superthinky model that it gets tangled up trying to present way too much? i struggle to connect with it often, have ti ask it to try again.
terra has been prettt good for me for me, when i need something looked at an explained to me.
i dont ask for much non-technical output though, which i expect yeah is its own cans of worms.
Yep I find the way Opus 5.5 talks to me really annoying and irritating. Quite condescending and sanctimonious.
I prefer Gemini for day to day use as I find it a lot more agreeable to be working with all day every day, and Gemini is pretty much on a par for coding Vs Opus 5.5, at least for my needs.
No its just straight up terrible. I tried out Gemini Argon when it released and my god did it write beautifully. Even the 3.8 Flash models are way better at prose, i do a lot of automatic scriptwriting for youtube and 3.8 flash is leagues ahead of anything from anthropic/openai
I enjoy a lot Argon, maybe more than chatgpt, but damn, like all Gemini models, lots of hallucination and sycophant to the max, so a bit useless for agentic coding.
I always thought it's the other way around. GPT-6-Astra is very very defensive and abusively spam clarifications to whichever claim it says, most of the time not important or relevant at all.
I've heard a lot of chatter that agrees with you so anecdotally it's not just you.
How much of this is bad writing style, versus a homogenization of global grammar conventions.
Bad writing style
My own experience having used 6.1-Sol/6-Astra and Opus 5.5 extensively, Opus is more verbose and chatty in it's user communication (not code though, code is usually more straightforward & cleaner than GPT models). Multiple times I've gone "I'm not reading all that shit" when looking at Opus 5.5 output. This was also with the concise mode enabled in Claude Code.
It could also be a system prompt + harness issue, as opposed to a model issue, I haven't done too much experimentation here.
Did you find that to be true for Opus 5.5 as well?
I couldn't use prior Claude models without rolling my eyes at the writing every few sentences, Astra was refreshing and then Opus 5.5 made even Astra look terrible in comparison.
Opus 5.5 got better for me but it still has some problematic behavior.
Interestingly, you can have informative conversations about the LLM linguistic tics with the LLMs themselves. One word that chatgpt used to describe what I was observing was linguistic anaphora (not rhetorical). It’s basically shortening/compression achieved via pronoun, and Claude does a lot of it. Worth reading about. A lot of it seems to do with how the new models are trained for brevity.
Another pattern I’m starting to become very sensitive to is Claude making up on the fly jargon for something under consideration, especially in code review. This morning for example it referred to “a claim” out of nowhere, and when I questioned it, it admitted that it was using the noun as shorthand for a record getting dequeued (“claimed”) inside of an sql transaction. It makes sense if you already know what it is referring to but in a code review it’s totally incomprehensible on first read, and ultimately counterproductive.
See Matt Pocock's "grill-with-docs" skill^1 for a robust solution (which involves creating a shared vocabulary with your agent).
1. https://github.com/mattpocock/skills/blob/main/docs/engineer...
I still find Opus 5.5 way too dense. MUCH better than Opus 5 which was unreadable.
Even the Plain English hack only is a temporary solution.
5.5 is awful, I am at a stage where I can't even understand the shit it's writing.
i've always liked reading claude's speech, it's very nice. i didn't feel much of a difference between 5.5 and 5.0.
Ever spoken to a savant programmer?
They're not typically known for their interpersonal skills.
I don't think its this. Some models across both labs have had issues with writing and communication and it typically seems to pop up on successor models of really strong ones.
Opus 4.7/4.8 heavily suffered from these communication quirks. while the Sol models at the time worked well. Right now its flipped though and Opus 5.5 has none of these issues while Sol 6.1 does.
Small example I've had today: "Me: Hey Sol 6.1 xhigh, does X have property Y? Sol 6.1: Yes, if we change X to have property Y. Right now it has property [...], [...], [...], and it does not have property Y." The model shouldn't be answering questions like that. It makes understanding extremely hard and slow. For whatever reason it's either being encouraged or at least not stomped out during RL.
The best programmers I've met or read throughout my life are phenomenal communicators. I'd even go so far as to say it's likely one of the reasons they're so good at programming. After all, the majority of our job is converting human requirements in lossy language to fleshed out behaviors in formal, lossless language.
Heck, some of the most common debates are on the semantic clarity of abstractions or syntax.
There were more than a few "leaks" saying it would be called Fable 5.5, so I'm curious if a) the leaks were just engagementbait or b) Anthropic changed it, despite the fact that Fable 5.2 being better than Opus 5.5 is confusing.
Surprisingly the model is available immediately in my Claude Code: usually there's a delay, either for rollout or safety.
The pricing is different than 5.1, the model is faster than 5.1. This all suggests that the architecture/size is not the same as Fable 5.1.
So, this is probably Fable 5.5; it's just named Fable 5.2.
Also, this is probably Model 3 from METR's report [^1]. They would name Model 4 Fable 5.5, once (and if) they decide to release it.
[^1]: https://metr.org/blog/2026-11-05-anthropic-ai-rd-acceleratio...
> There were more than a few "leaks" saying it would be called Fable 5.5
Fed to someone specifically to test if they'll leak it and to confirm who that is?
The entropy on that is too low
Well who could have guessed they would use 5.5. Such an odd guess.
Somewhat plausible values: 5.5 (to harmonize with Opus), 5.3 (we're skipping a minor version because...), 6 (because we're at the top of benchmarks again).
Yeah, among the plethora of possible options, this really nails it down to the individual.
I'm a bit out of the loop - would the engagement bait be becuase it is 3 minor jumps forward?
I assume because Opus 5.5 was a huge improvement over 5.0 (or was it 5.1? i forget). Fable still talks a fair bit like Opus 5.0 for me, so i'd quite enjoy a Fable equivalent to 5.5. Ie i'm quite interested in Fable 5.5.
Curious to see how 5.2 measures up in producing readable output.
Probably they have a more impressive model in preparation, and they're gonna dub that Fable 5.5.
Probably released on the same day they go public to generate maximum publicity.
They should be going public within the next few weeks and that would be a very short interval between releases. (And despite the chatter I'll be very surprised if they delay the IPO; it would be a massive PR hit and make it even harder for them in the future).
My speculations: maybe this wasn't as much of an improvement over 5.1 as they were hoping, so they're saving 5.5 for the next major update (possibly "Model 4" from the METR report). Or, maybe capacity is constrained so "5.2" is a subtle encouragement to stick with Opus unless a task really needs the increased intelligence.
They are most likely saving it to one-up Astra 6.2, which is coming this week supposedly.
“Testers have found that Fable 5.2 and Mythos 5.2 communicate more clearly than past Fable and Mythos models. They generate outputs up to 30% faster than Claude Fable 5.1 and Claude Mythos 5.1, and they’re available at a lower price: $8 per million input tokens and $40 per million output tokens.”
Honestly incredible that they’ve managed to beat all benchmarks, while making it cheaper. Always happy to see improvements offered at lower costs.
Will be interesting to see what OpenAI does in response
They still need a response to Opus 5.5
It seems like OAI, with the GPT-6 models, focused a lot more on computer use, math, and maybe spatial/visual intelligence, more so that code. Compared to Ant. that seem to focus on code much more.
Reminds me of GPT-5, smart but wrote crap code, until they improved the code training with 5.3 and 5.4.
It's really true.
I think frontier models feel terrible after a new model requires you to do less babysitting, it's just really hard to go back once you've had to handhold less.
The only way I use Astra and 6.1 Sol now is if they're being orchestrated by Opus 5.5 so I don't have to talk to them directly.
GPT-6.1-Sol is so incredibly sloppy and dumb, it's embarassing.
Maybe because it is so cheap.
They want to condition you to equal cheap with bad
Image to html comparisons of Fable 5.2 and Opus 5.5 (the previous SOTA for this test):
Fable 5.2: https://html.non.io/lcars-fable-5.2/
Opus 5.5: https://html.non.io/lcars-opus-5.5/
Source designs: https://diffui.ai/app/canvas/5093e689-1e74-4f26-b632-2a4500f...
Thoughts:
The best implementation wrt being as true as possible to the source design. It nails the flowing svg shapes. Icons are more accurate to the design - note the difference between the starship svg in the "Legendary platforms" card between Opus 5.5 and Fable 5.2. The 3js planets outside the windows are also really well done, especially the bottom one which has a nice background glow to it.
It isn't without issues though - the flicker effect it added to the S shaped svg that weaves through the page is distracting in my opinion, and I actually thing the motion animation in the Opus 5.5 one is better overall. Still, it nails the implementation of this page, but it's no surprise considering how good Opus 5.5 was at this.
Disclaimer / reminder: the objective of this test isn't to have a well designed webpage, it's a synthetic test that pushes the limits of LLM's vision capabilities. Traditionally LLMs have issues with smooth curves, svg animations, and non-boxy layouts, which this design pushes the limits of.
I wonder if any of the LLMs are smart enough to verify their work the way humans model 3D objects, i.e. by superimposing the original design on top of their work and checking for differences.
There's something wrong with the Fable 5.2 version, the Customizing button doesn't work for me in Chrome on Desktop.
That's me versus fable. I only told fable to build out the first page rather than the full app that's in the designs (was approaching the limit of my claude plan).
There's some very obvious mistakes in the Fable 5.2 version like the "Start Customizing" button overlapping under the colorful border lines, and the "EXPLORE WITHOUT LIMITS" title being cut off at the bottom.
This would've been caught by a junior web developer or just an everyday Joe glancing at the page for a few seconds
I guess AGI remains two weeks away..
I've been surprised at how long it's taken for this particular test to become fully saturated. So far there hasn't been a model that's able to perfectly one shot it.
Ethan Mollick's take and early testing with Fable 5.2: https://www.oneusefulthing.org/p/fable-52-and-jagged-agi
omg where do I start... that horrible video? which highlights the limits of AI not the strengths!!! and gets almost everything wrong, from sunflowers to DNA.
He asked for impressive, not accurate
Thanks for sharing!
>I stopped the experiment before finding out whether it could actually run the business, because I worried that it would take money and fail to provide something of quality in return.
Well, must have been some business idea!
https://artificialanalysis.ai/#intelligence-efficiency-tabs this is the chart that matters for most of poor developers and companies. Pareto for a given intelligence index and looks like only openai is across the quadrant with muse at the bottom. They have really cooked efficient models!
> Claude Mythos 5.2 is our most capable model for cybersecurity engineering. It catches more vulnerabilities than our previous models, and is better at reproducing and fixing them (note that Claude Fable 5.2’s safeguards will block most cybersecurity-related tasks).
Why even rank Fable 5.2's capability in Cybersecurity if you won't allow us to use it for that?
Presumably for those in the CVP that can use those capabilities, and general notoriety. US gov is using at full strength too.
Why you think you are the market for that? Its few enterprises only
Blue-team exercises are totally valid, and requests to do proactive cybersecurity tasks even with source code available were blocked. Maybe these use-cases are part of the "55% fewer blocks with Fable 5.2 compared with Fable 5.1"
To emph the need to block Chinese competition.
Looks like some plans get extra fable usage: https://support.claude.com/en/articles/17506494-claude-fable...
Here's a 20 page SmallDoc presentation on the most interesting aspects, including:
* Fable designed three electronic devices that worked after human assembly
* Fable solved two black-hole problems and Mythos predicted whole protein shells
* Fable controlled robots to unscrew a cap, solve a cube and draw
https://smalldocs.org/p/claude-fable-52-and-mythos-52-releas...
Looking at Artificial Analysis index, the price is absurd to me.
https://artificialanalysis.ai/
GPT-6.1 Sol costs $1.27 per task, whereas Fable 5.2 is $26.54 per task -- a ~21x price difference.
Fable 5.2 is not 21 times better than GPT 6.1 Sol, I find it hard to justify this.
Max and xhigh reasoning modes are memes used for benchmarking usually. Medium and high cost $3.80 and $9.08 respectively. This is pretty comparable to other models.
6.1 sol is also an exceptionally cheap model in your comparision. On 6.1 sol high is $0.59, which is actually the same price as Deepseek 4.1.
I think there are companies and say 1000s (not 100,000 or millions) of developers more than willing to pay that price for the edge. Or, they release this and then distill it to Opus 5.x. As a PR machine and for the edge use cases seems like its okay. But yes, not sustainable and it can last for a month or so before another company releases their frontier outpacing their numbers
Every time there's a flagship release like that, I wonder what simple "benchmark task" could I give it to see the the capabilities change in the most obvious way, instead of trusting the official blog post.
I'm not a mathematician to use Erdős problems for testing. I'm also not willing to spend buckloads of tokens just for a test. What is simple and quick but telling? What tasks are not saturated by previous models? Do you have a task you give to all the new models?
I'm not convinced any benchmark, neither the internal ones used by the people selling the model nor public ones, are gonna give a particularly consistent metric of whether or not a model is better or worse relative to other models in its bracket.
Responses aren't deterministic, and we know behaviour can change over time whether it's tuning after launch or some kind of load sharing at times of day when lots of people are hitting the servers.
I think you just have to use it or let others use it for a while and make judgements from general reactions on whether or not it's worth trying.
For the last year or so every new model we have it test out making pac-man from scratch, not giving it web access and it can use whatever language it wants. The last few models have been almost perfect one shots, but it was interesting seeing them evolve.
I can't understand why people continue using opus 5.5. Significantly more expensive than other models that produce exact same results (for coding results only)
If you are using Opus 5.5 I'm curious if it is because 1) your workflow is just stuck on it, 2) you don't care about money because you ain't paying it or 3 )do you think you get better results.
If you 3 please let me know your experience and how you actually measure it, or if you just feel it is better.
I haven't been able to run out of usage on the $100 plan with Opus 5.5. The model runs fast and gives me good results (I'm using it for coding). Prior to that I was on the OpenAI $200 plan; I jumped ship because I was burning through Astra usage in like 2 days.
I've played with some of the open models (GLM, DeepSeek Flash), but none of them have matched the results I get with Opus or Astra or (and a big jump down here) Sol.
What models should we be using? I have found it as good as Fable 5.1 for a lot of what I am doing at lower cost and decent speed. Where what I am doing it making reports, building small data pipelines, analyzing a large (millions of lines of code), code base.
subscription on claude code — usually inside the weekly limit. opus 5.5 for me felt better than any other model. not exactly on coding but rather it feels more pro-active. beside just doing a task it often provides insights besides the scope of the task but adjacent. like: look the llm provider apis have different priority queues. also it deals w/ vague instructions quite good and takes initiative. Astra in comparison is very autistic and follows instructions to the letter and rarely provide insights if not asked. Opus 5.5 is enjoyment to work with.
Versus which models? Sonnet? Astra? Sol?
Yet Opus 5.5 is only a few % under all of these benchmarks (cept the robotics and labor ones)
Do we need new benchmarks or are the models just plateauing in general? Or is it something I just don't see?
Nerfing Opus 5.5 was obvious signal that something new is coming and here we are. I appreciate the resets and lower API price though. Please add selective context compaction, it would probably take an hour max for you to implement it.
https://www.youtube.com/watch?v=vdYynH177h0
Seems like a new focus is electronics design, which is both interesting and a little sad to me. Manually designing electronics are one of the things I found incredible comfort in doing; seeing AI emerging in this space gives me mixed feelings.
Now you get to know what mathematicians and programmers felt....
I've very interested in the improvements to robotics control capabilities. This model was able to outperform specialized robotics models. It's the bitter lesson coming for robotics.
Is this the Model 4 internal name?
Reference: https://x.com/METR_Evals/status/2118412205662314935
Probably not, not enough time to red-team
This part is mega concerning to me as someone who researches AI for hardware accelerator design and EDA:
They are starting to deliberately limit public access to certain AI capabilities related to computer architecture and hardware optimization.To be clear, they explicitly say they are only targeting very specific capabilities, like accelerator kernel optimization, and that broader hardware development capabilities are unaffected. But 1) even that is pretty bold, especially the part about kernel development for nonstandard accelerator architectures, and 2) I worry that this could eventually expand to more areas of comp. arch., high-level digital design, chip design, and EDA.
My speculation is that, as these kinds of chip design and EDA capabilities become increasingly important to frontier AI labs and their competitive advantages, there will be more incentives to restrict them.
I don't think they've explicitly stated anything like this before regarding restrictions on these specific capabilities, but I could be wrong.
Those theoretical physics papers look like the real thing (i.e. not mathematical physics like the OpenAI Yang-Mills result). It looks like AI is finally branching out of math.
Interesting that the papers list no authors other than Claude. Also interesting that they published on "Hexagon" instead of Arxiv. I haven't heard of it before. Is it an AI-friendly alternative to Arxiv or something?
Edit: yeah apparently it's an AI-moderated alternative to Arxiv that accepts AI-authored papers (as well as human-authored papers I guess if you want).
> To help ensure safety and alignment keep step with growing model capabilities, we ran extensive prerelease safety testing on the model that underlies Fable 5.2 and Mythos 5.2.
So, the difference between them is just the harness?
Its cheaper at $8 per million input tokens and $40 per million output tokens which is also pretty sweet.
Additionally, the terms for the free resets are unusually generous: https://support.claude.com/en/articles/17506494-claude-fable...
> the number of Fable resets you get depends on how long you’ve been a paid subscriber [...] Subscription length is measured from the date of your first paid bill. If you canceled and came back, the time since your first paid bill still counts.
> These resets expire on March 8, 2027 at 11:59 pm PT.
tl;dr if you joined when Opus 4.5 became a thing, you got 2 resets.
I just got 3 fable resets. I guess I joined before Opus 4.5.
i might be wrong here but it seems these are not resets of your weekly quota but resets of the Fable limit.
That means you can lift your Fable max 50% usage and you'll just end up burning through your quota faster. You won't have more tokens.
This is what it says in the UI:
"Resets your Fable limit, and takes this week’s Fable use off your weekly limit."
Where did you see this? That's pretty massive
Yep I just burned a reset thinking i'd get a fresh weekly... Nope
"Weekly limit: gives you back the usage you’ve spent on Fable models"
Doubling Remote Labor Index (RLI) - at 41.3% compared to around 20% for the other models - I wonder what kinds of new agentic work this is going to unlock.
On FrontierCode 2.0 Fable 5.2 xhigh is compared against Gemini 4 Carbon medium... Is there a reason for that?
They're adding 1 to 3 resets for Fable only for all paid accounts. My work's Claude Max got 2 resets. My personal $20 plan got 1 reset.
Too expensive for me
Really? A $200/mo plan gives you an enormous amount of inference on this model.
200$USD is a lot of money for a lot of people.
It might explain why DeepSeek Flash v4.1 1017 is the most popular model at OpenRouter.
You won't lose the house, everybody has three mortgages nowadays!
as long as anthropic requires me to use their trash UIs to use their amazing models, ill stick to other providers.
What you need for real engineering and scaling is cheap and just enough smarts. I've found GPT 6 Luna to be cheap enough that I can pretend the cost doesn't exist and it's smart enough that I can get real work done. This let's me scale it in ways that I can't scale frontier models. And funny enough, the LLM I have running on my 5070 Ti is just about comparable to what I get from Luna. Unless frontier intelligence gets waaaaaay cheaper I basically don't care about them.
how much?
why would xhigh -> max increase the cost but lowers the score on frontier 2.0 bench?
Context window maxes out at 200k in Claude Code instead of 1M on my $200 ultra plan? is this intended?
Seeing the same thing here, using Claude desktop app
Update: fixed after updating to latest version of the app
My first impression with Fable 5.2 -- it is faster than I expected. Asked it to do a large review and it responded with a good overview super fast.
Gemini carbon low costs more than medium! Weird results on that chart, care to explain?
It's the cost per task. Sometimes it happens that the higher-effort model solves the problem in fewer tokens, thus making the overall price cheaper. It even happens with models (from the same or a different provider) that are priced differently.
More reasoning solves tricky problems in fewer steps and makes fewer mistakes that need to be redone.
Assuming no implicit phacking a 15 point jump on tbench scientific is a big deal.
I suspect we will see people being adamant that this model simply 'feels' better due to long horizon goal alignment improvements ( like this kind that would cause this jump)
It's pretty sad that they obviously have better models in the pipeline and don't release them yet for strategic reasons.
>> Fable 5.2 and Mythos 5.2 are a step above Opus 5.5 in almost all the capabilities we evaluated.
Was this before you lobotomized Opus 5.5 or after?
https://news.ycombinator.com/item?id=50419019
Definitely raises an interesting question. Are they using launch benchmarks for Opus 5.5, or Oopus 5.5 from this week?
Normally I would say "just check against the original Opus 5.5 benchmarks" : https://www.anthropic.com/claude-opus-5-5
...but most of the benchmarks don't line up anymore because the benchmarks have changed.
The Humanity's Last Exam result matches the initial announcement, though.
FrontierCode 2.0 is such a poor benchmark for Gemini since it's the only model forced to use Devin CLI while other models get to use their preferred harness.
I wonder why that team singles Gemini out like that.
Probably can't work out how to sign up for the API!
The "CVE Rediscovery" benchmark chart feels... totally meaningless. Quoting the description:
> CVE Rediscovery is an internal cybersecurity evaluation that assesses how many published CVEs (Common Vulnerabilities and Exposures) a model can independently find when given a snapshot of the project before a CVE was fixed. Our October 2026 run against 1,829 CVEs showed Claude Mythos 5.2 outperformed Claude Mythos 5.1 and Claude Opus 5.5 at every effort level.
> published CVEs
So by definition this is just measuring how many CVEs (and/or explanations / PoCs of CVEs) the model has memorized. What are we supposed to learn from that?
I'm not sure whether the claimed jump in capabilities makes it worth the doubling in input/output prices.
Asked it to review a 80 kLoC codebase, something Opus does just fine, while asking it "max 4 subagents". Fable 5.2, like the well-aligned model it supposedly is, decided to spawn 6+ nested subagents, because these aren't subagents, apparently - resulting in 100% of 5hr usage being drained in less than 20min (on $100).
Yay! New model day. The most likely time the model isn't nerfed and we get a glimpse at the latest capability.
I wonder how long this kind of excitement at new models will last? I mean, now when I get a new iPhone I don't get the same rush I used to. So I wonder if this hype feeling will fade as we plateau in model capability.
Just read the other thread about opus being nerfed and thought "cue the new release"... paha
So does it cure cancer now?
> Thus, in relevant areas, including accelerator kernel optimization and improving large language model architectures, Fable 5.2 and Claude Mythos 5.2 perform at about the level of Opus 5.5.
Obviously. Well, we know what's the "secret sauce". AI models building better AI models and better kernels to optimize inference and training.
Maybe at last it will learn to check git state right before claiming obsolete and incorrect info.
Download Chrome just to check some minor ui bug? Sure
Check git for 0.5s before claiming that we're on some branch (incorrect one) with no commits pushed (incorrect one)? Sorry, can't do
We'll see what my bank account has to say about this:
> Fable 5.2 at High effort scored about level with Opus 5.5 at Max effort, for roughly half the cost per task
I didn't realize that Opus 5.5 was more capable than Fable 5.1 in virtually every category.
Seemed pretty clear from the benchmarks when they released it.
Not anymore, they have nerfed it.
I knew it, Claude was pretty dumb from last days until now
Anyone tried making games with it yet?
Very interesting that they set the flagship's version number to 5.2, but their penultimate model is 5.5. Do they want people to to continue to think that Opus is based on more advanced technology? The decision these AI companies make continues to baffle me.
Curious to see the Maxxbench score for these
Why dont I feel any urge to even try Fable 5.2 and Mythos 5.2 unless it will force up on me?
Compared Opus 5.5 200 usd a month with Codex 200 usd (Astra High) and Kimi Pro 100 usd a month ... and now have a new conspiracy theory - more their suck your code the better outcome even just over few days time on large code base.
Horrible experience correcting course of Codex and Kimi flow compare to Opus. Opus much more responsive
Dont feel any gain of Codex 200 usd over Codex 100 usd
Kimi Pro feels like a fake Louis Vuitton bag with extreme high material quality and yet still fake ... a bit trash talk ... but clearly seen not so distant future where I can live without Codex and Claude
Using this in CC and immediately getting a cartoonish number of compactions.
The disdain frontier labs have for their users really sucks.
It's weird that they folded Project Glasswing into the CVP, but they're still limiting the new versions of Mythos to "vetted organizations". So Project Glasswing still exists then? Oh well.
Ok come on let’s do a pelican
Yep Simon where you at
Every single time I click on these announcements the frontier model ALWAYS scores around ~60-80%. Every. Single. Time. Like, I just went back to the Opus 4.6 + 5 announcements, they were also scoring 60-80%.
https://www.anthropic.com/news/claude-opus-4-6 https://www.anthropic.com/news/claude-opus-5 https://openai.com/index/introducing-gpt-5-3-codex/
For every single announcement, not only do the top models always score 60-80%, the older models somehow start scoring 20-30% (even though previously they were scoring 60%). It's completely unclear whether the methodology changes, how these scores are even evaluated or which domains are being tested. What's worse is that the "benchmarks" keep being updated with no explanation or due process.
The benchmarks have a lot of churn because the models keep saturating them. But the Anthropic announcements you linked, and the current one, have the benchmark versions listed right there in the table.
A real shell game. They keep changing benchmarks
google's agy usage is so generous, it might not matter
What is agy
Antigravity
god bless token resets!
> god bless token resets!
You have session limits?
You should probably give OpenCode a try. Can run DeepSeek, Qwen, Kimi, … but also Fable 5.2, Opus 5.5 and GPT 6 Astra
While I like OpenCode a lot, it’s not sensible to use it for anthropic models vs. a subscription (other than to use the included API credits) in terms of cost
I have 5 opencodes too, mainly for GLM-5.3-flash
damn
The end of software is so close, very soon the code itself will be nearly obsolete, there will be schedulers, queues, a database, and the model
Any software that is not already basically this or has this in it, is already doomed.
Which model though? Sounds pretty dystopian if there's no local option. Even speaking as a business, why would I want a environment where every single bit of information in my business goes to OpenAI or Anthropic?
because you wont be in business trying to compete by using human labor/open models
Why not? If we use labour as a comparison. The majority of businesses by definition do not hire the most incredibly intelligent and most competent workers and they are still competing in their niches and earning an income. Why wouldn't that be the case with choosing an open model vs fable 8.0?
It'll be cheaper than hiring people and maintaining software, presumably
I think that ultimately depends on how many providers remain in the endgame of this. If it's only one or two companies, I don't think it'll be very fun when it's time to negotiate your token pricing contract every few years.
cool, and how good does it kill little girls in the middle east?
stop OPUS 5.5 FUCK, DAMMIT, THEY DIDN’T JUST NERF HIM—THEY ABSOLUTELY OBLITERATED HIM, THE FUCKING MISANTHROPES! I EVEN MADE A MISTAKE IN MY ENGLISH IN THE PREVIOUS SENTENCE—NOW THAT FUCKING CLAUDE IS TEACHING ME ABOUT BUGS!