OpenAI's GPT-6.2 Astra and Sol: 5.2-point benchmark gain at less than half the per-task cost
YashasGunderia · x · 2026-11-18
reachvb announced GPT-6.2 Astra and Sol; YashasGunderia quote-shared it noting his usage limits were already 100% consumed.
Key details:
- Positioning: Astra targets the hardest projects; Sol delivers strong results at a fraction of the cost;
- Capabilities: cleaner code and frontends, template-following slides and spreadsheets, stronger scientific reasoning;
- Benchmarks: on Agents' Last Exam, 6.2 Astra at xhigh scores 5.2 points above GPT-6 Astra at max, at less than half the per-task cost; Sol nears flagship level on FrontierCode 2.0.
Related event: OpenAI Launches GPT-6.2 Astra and Sol Dual Models(32 posts)→