Paweł Huryn's 105-bug test puts GPT-6.2 Astra first at 62.7 for $28.35
Paweł Huryn says he ran models on two repositories with 105 replanted bugs and asked them to find and fix what they could. He reports GPT-6.2 Astra as the new #1, and calls it strong at code review.
His scores, with the cost of the run:
- GPT-6.2 Astra (max): 62.7 for $28.35
- GPT-6.2 Astra (xhigh): 60 for $23.53
- Claude Fable 5.2 (max): 58 for $139.65
- Gemini 4 Carbon (high): 57.3 for $31.32
- Claude Sonnet 5.5 (max): 51.3 for $154.03
- GPT-6 Astra (max): 45 for $33.03
That is 17.7 points more than GPT-6 Astra at max effort and 17 at xhigh. Huryn says GPT-6.2 Sol and more effort levels will follow in the thread. The results are his own and one test; the post had 1,343 points on X by November 18.
Sources
Get news like this as a push.
AI Tracker alerts you when something changes for the tools you use.