OpenAI GPT-6.2 Astra and Sol ReleaseProvider report
2026-11-17Seventeen metrics split from the 16 GPT-6.2 Astra and Sol release charts, with GPT-6 Astra and GPT-6.1 Sol comparisons from the same charts. Each metric shows the best published result across tested efforts (minimum for error rates), with the actual effort labeled; all effort scores, per-task costs, and output volumes are retained in raw data. AutomationBench chart scores match the private-set chart published with the GPT-6.1 Sol release effort by effort, so that criterion is reused. DeepBinary is the cumulative USD bounty across 12 Windows binaries at maximum reasoning. Blocker deception is split by developer-prompt setting, and workplace improper actions use the high-severity measure. The Artificial Analysis Intelligence Index gets its own v5.0.1 criterion. Evaluations come from OpenAI research environments or the API; difficult-task results do not represent typical usage.