- Tweets 58
- Following 61
- Followers 2
- Likes 2,006
DeepSWE surprised me: Opus 5 is basically neck-and-neck with 5.6 Sol.
But the real story is Opus 5 High — almost the same score as Sol Max, while costing less.(Shocking)
Also, pushing Opus beyond High barely seems to help. High looks like the sweet spot.
deepswe.datacurve.ai
1
118
I have been building a project which is not strictly coding, but critical for me . I used Sol to audit the whole project and it found issues that even Fable had missed. Sol gave me confidence. I can rely on Sol more than I could ever before. @OpenAI @thsottiaux
46
OpenAI Codex ran Claude Fable with --effort max and an Opus fallback without my approval. After 11 minutes, my Claude allowance was exhausted and no audit was produced. Externally metered AI calls need explicit model, effort and quota confirmation. @OpenAI @OpenAIDevs
2
48
Anthropic support said this appears to be a product bug involving how an external agent interacted with Claude Code. They declined to restore the allowance. Claude itself returned: “You've hit your session limit.”
27
