They trained it in 33 days for ~20m (that includes apparently not only the infrastructure but also the salaries over a 6 month period). And the model is coming close to QWEN and Deepseek. Pretty impressive
The price/scaling of training another same class model always seems to be dropping through the floor but training models which score much better seems to be hitting a brick wall.
E.g. gemini-3-pro tops the lmarena text chart today at 1488 vs 1346 for gpt-4o-2024-05-13. That's a win rate of 70% (where 50% is equal chance of winning) over 1.5 years. Meanwhile, even the open weights stuff OpenAI gave away last summer scores between the two.
The exception seems to be net new benchmarks/benchmark versions. These start out low and then either quickly get saturated or hit a similar wall after a while.
> E.g. gemini-3-pro tops the lmarena text chart today at 1488 vs 1346 for gpt-4o-2024-05-13. That's a win rate of 70% (where 50% is equal chance of winning) over 1.5 years. Meanwhile, even the open weights stuff OpenAI gave away last summer scores between the two.
Why do you care about LM Arena? It has so many problems, and the fact that no one would suggest using GPT-4o for doing math or coding right now, or much of anything, should tell you that a 'win rate of 70%' does not mean whatever it looks like it means. (Does GPT-4o solve roughly as many Erdos questions as gemini-3-pro...? Can you write roughly as good poetry?)
It'd certainly be odd if people were recommending old LLMs which score worse, even if marginally. That said, 4o is really a lot more usable than you're making it out to be.
The particular benchmark in the example is fungible but you have to pick something to make a representative example. No matter which you pick someone always has a reason "oh, it's not THAT benchmark you should look at". The benchmarks from the charts in the post exhibit the same as described above.
If someone was making new LLMs which were consistently solving Erdos problems at rapidly increasing rates then they'd be showing how it does that rather than showing how it scores the same or slightly better on benchmarks. Instead the progress is more like years since we were surprised LLMs were writing poetry to massage out an answer to one once. Maybe by the end of the year a few. The progress has definitely become very linear and relatively flat compared to roughly the initial 4o release. I'm just hoping that's a temporary thing rather than a sign it'll get even flatter.
Frankly, as a neutral 3rd party, this is a lot of words that amount to an excuse for using only LMArena, and the rationale is quite clear: it’s for an unrelated argument that isn’t going to ring true to people, especially an audience of programmers who just spent the last year watching the AI go from being able to make coherent file edits to multi hour work.
LMArena is, de facto, a sycophancy and Markdown usage detector.
Two others you can trust, off the top of my head, are LiveBench.ai and Artifical Analysis. Or even Humanity’s Last Exam results. (Though, frankly, I’m a bit suspicious of them. Can’t put my finger on why. Just was a rather rapid hill climb for a private benchmark over the last year.)
FWIW GPT 5.2 unofficial marketing includes the Erdos thing you say isn’t happening.
```During development of the RSDB, we noted significant enough performance gains from it that we decided to integrate it during phase 3 of the Trinity Large training run instead of waiting for a later training run. While the
data distributions between phase 2 and phase 3 make direct comparison difficult, the overall effect was notable: BatchHet reduced by a factor of 4.23x, and step-to-step variance reduced by a factor of 2.4x (see Figure 1), a significant improvement when compared to the default packing strategy. We note that training runs without the RSDB exhibit much higher values in the higher-order moments of the running loss distribution, which we believe to correlate with network instability during training.
```
Page 9 of the technical report has more details, but it looks like they found some data prep methods as well as some other optimizations that overall worked out really well. I don't think it was any one particular thing.
As far as Llama 4 goes, it was only referenced as a similarly sized model, they called it one of their model "peers"; I don't think they intended any sort of quality comparison. Llama 4 was notable for sparsity, despite its poor performance and reception, some of the things they achieved technically were solid, useful research.
you can’t directly compare losses because they changed the data distribution for each phase ( I think. 100% guaranteed they change the data distribution after the 10 trillion token mark, that’s when they start adding in instruction following data, but I don’t know for sure if the other phase changes also include data distribution changes.)
Given that it's a 400B-parameter model, but it's a sparse MoE model with 13B active parameters per token, would it run well on an NVIDIA DGX Spark with 128 GB of unified RAM, or do you practically need to hold the full model in RAM even with sparse MoE?
Even with MoE, holding the model in RAM while individual experts are evaluated in VRAM is a bit of a compromise. Experts can be swapped in and out of VRAM for each token. So RAM <-> VRAM bandwidth becomes important. With a model larger than RAM, that bandwidth bottleneck gets pushed to the SSD interface. At least it's read-only, and not read-write, but even the fastest of SSDs will be significantly slower than RAM.
Can run with mmap() but it is slower. 4-bit quantized there is a decent ratio between the model size and the RAM, with a fast SSD one could try to see how it works. However when a model is 4-bit quantized there is often the doubt that it is not better than an 8-bit quantized model of 200B parameters, it depends on the model, on the use case, ... Unfortunately the street for local inference of SOTA model is being stopped by the RAM prices and the GPU request of the companies, leaving us with little. Probably today the best bet is to buy Mac Studio systems and then run distributed inference (MLX supports this for instance), or a 512 GB Mac Studio M4 that costs, like 13k$.
So refreshing to see open source models like this come from the US. I would love for a 100Bish size one that can compete against OSS-120B and GLM air 4.5
With such a high throughput because of sparsity, I'm particulary interested in distilling it into other architectures. I'd like to try a recurrent transformer when I have the time
They trained it in 33 days for ~20m (that includes apparently not only the infrastructure but also the salaries over a 6 month period). And the model is coming close to QWEN and Deepseek. Pretty impressive
The price/scaling of training another same class model always seems to be dropping through the floor but training models which score much better seems to be hitting a brick wall.
E.g. gemini-3-pro tops the lmarena text chart today at 1488 vs 1346 for gpt-4o-2024-05-13. That's a win rate of 70% (where 50% is equal chance of winning) over 1.5 years. Meanwhile, even the open weights stuff OpenAI gave away last summer scores between the two.
The exception seems to be net new benchmarks/benchmark versions. These start out low and then either quickly get saturated or hit a similar wall after a while.
> E.g. gemini-3-pro tops the lmarena text chart today at 1488 vs 1346 for gpt-4o-2024-05-13. That's a win rate of 70% (where 50% is equal chance of winning) over 1.5 years. Meanwhile, even the open weights stuff OpenAI gave away last summer scores between the two.
Why do you care about LM Arena? It has so many problems, and the fact that no one would suggest using GPT-4o for doing math or coding right now, or much of anything, should tell you that a 'win rate of 70%' does not mean whatever it looks like it means. (Does GPT-4o solve roughly as many Erdos questions as gemini-3-pro...? Can you write roughly as good poetry?)
It'd certainly be odd if people were recommending old LLMs which score worse, even if marginally. That said, 4o is really a lot more usable than you're making it out to be.
The particular benchmark in the example is fungible but you have to pick something to make a representative example. No matter which you pick someone always has a reason "oh, it's not THAT benchmark you should look at". The benchmarks from the charts in the post exhibit the same as described above.
If someone was making new LLMs which were consistently solving Erdos problems at rapidly increasing rates then they'd be showing how it does that rather than showing how it scores the same or slightly better on benchmarks. Instead the progress is more like years since we were surprised LLMs were writing poetry to massage out an answer to one once. Maybe by the end of the year a few. The progress has definitely become very linear and relatively flat compared to roughly the initial 4o release. I'm just hoping that's a temporary thing rather than a sign it'll get even flatter.
Frankly, as a neutral 3rd party, this is a lot of words that amount to an excuse for using only LMArena, and the rationale is quite clear: it’s for an unrelated argument that isn’t going to ring true to people, especially an audience of programmers who just spent the last year watching the AI go from being able to make coherent file edits to multi hour work.
LMArena is, de facto, a sycophancy and Markdown usage detector.
Two others you can trust, off the top of my head, are LiveBench.ai and Artifical Analysis. Or even Humanity’s Last Exam results. (Though, frankly, I’m a bit suspicious of them. Can’t put my finger on why. Just was a rather rapid hill climb for a private benchmark over the last year.)
FWIW GPT 5.2 unofficial marketing includes the Erdos thing you say isn’t happening.
What did they do to make the loss drop so much in phase 3?
Also, why are they comparing with Llama 4 Maverick? Wasn’t it a flop?
```During development of the RSDB, we noted significant enough performance gains from it that we decided to integrate it during phase 3 of the Trinity Large training run instead of waiting for a later training run. While the data distributions between phase 2 and phase 3 make direct comparison difficult, the overall effect was notable: BatchHet reduced by a factor of 4.23x, and step-to-step variance reduced by a factor of 2.4x (see Figure 1), a significant improvement when compared to the default packing strategy. We note that training runs without the RSDB exhibit much higher values in the higher-order moments of the running loss distribution, which we believe to correlate with network instability during training. ```
Page 9 of the technical report has more details, but it looks like they found some data prep methods as well as some other optimizations that overall worked out really well. I don't think it was any one particular thing.
As far as Llama 4 goes, it was only referenced as a similarly sized model, they called it one of their model "peers"; I don't think they intended any sort of quality comparison. Llama 4 was notable for sparsity, despite its poor performance and reception, some of the things they achieved technically were solid, useful research.
you can’t directly compare losses because they changed the data distribution for each phase ( I think. 100% guaranteed they change the data distribution after the 10 trillion token mark, that’s when they start adding in instruction following data, but I don’t know for sure if the other phase changes also include data distribution changes.)
I'm particularly excited to see a "true base" model to do research off of (https://huggingface.co/arcee-ai/Trinity-Large-TrueBase).
Given that it's a 400B-parameter model, but it's a sparse MoE model with 13B active parameters per token, would it run well on an NVIDIA DGX Spark with 128 GB of unified RAM, or do you practically need to hold the full model in RAM even with sparse MoE?
Even with MoE, holding the model in RAM while individual experts are evaluated in VRAM is a bit of a compromise. Experts can be swapped in and out of VRAM for each token. So RAM <-> VRAM bandwidth becomes important. With a model larger than RAM, that bandwidth bottleneck gets pushed to the SSD interface. At least it's read-only, and not read-write, but even the fastest of SSDs will be significantly slower than RAM.
That said, there are folks out there doing it. https://github.com/lyogavin/airllm is one example.
Can run with mmap() but it is slower. 4-bit quantized there is a decent ratio between the model size and the RAM, with a fast SSD one could try to see how it works. However when a model is 4-bit quantized there is often the doubt that it is not better than an 8-bit quantized model of 200B parameters, it depends on the model, on the use case, ... Unfortunately the street for local inference of SOTA model is being stopped by the RAM prices and the GPU request of the companies, leaving us with little. Probably today the best bet is to buy Mac Studio systems and then run distributed inference (MLX supports this for instance), or a 512 GB Mac Studio M4 that costs, like 13k$.
Talking about RAM prices, you can still get a framework Max+ 395 with 128GB RAM for ~$2,459 USD. They have not increased the price for it yet.
https://frame.work/products/desktop-diy-amd-aimax300/configu...
Pretty sure those use to be $1999 ... but not entirely sure
Yep. You be right. Looks like they increased it earlier this month. Bummer!
The only thing I question is the use of Maverick in their comparison charts. That's like comparing a pile of rocks to an LLM.
There aren't too many base models out there to compare against.
What exactly does "open" mean in this case? Is it weights and data or just weights?
It's always open weights.
It's never open data
Well, it is, it's your data to begin with after all but admitting that would create some problems.
This model is sort of interesting since it seems to be using a lot of synthetic training data – but your point stands
So it's a rip off of a rip off, is that whats interesting?
> We optimize for performance per parameter and release weights under Apache-2.0
How do they plan to monetize?
So refreshing to see open source models like this come from the US. I would love for a 100Bish size one that can compete against OSS-120B and GLM air 4.5
This is a wonderful release.
Is anyone excited to do ablative testing on it?
With such a high throughput because of sparsity, I'm particulary interested in distilling it into other architectures. I'd like to try a recurrent transformer when I have the time