Cache hit rates of Inference are more meaningful than the headline costs

2 comments

• faangguyindia 7 hours ago

  How come dirac's advance code editing technique is not being by Claude or Codex and other agents?

  I still routinely see these agents throwing away work because the editing failed?

  https://github.com/openai/codex/issues/11601

  https://github.com/anthropics/claude-code/issues/25775

  It seems they didn't look into it seriously.

  [-]

  • cyanydeez 21 minutes ago

    opencode has a plugin for similar work: https://github.com/Opencode-DCP/opencode-dynamic-context-pru...

    It allows the model to remove context pollution.