GPT-6.2 Astra sets Terminal-Bench Science record: +9 points over GPT-6 Astra at under 1/3 cost

OpenAIDevs · x · 2026-11-18

OpenAI reports GPT-6.2 Astra sets a new high on Terminal-Bench Science 0.1, which tests scientific work in a terminal: 9 percentage points above GPT-6 Astra at under 1/3 the cost. The quoted prior tweet adds coding-efficiency data — on FrontierCode 2.0, GPT-6.2 Sol is within half a point of Astra at 1/5 the cost. Charts show Astra hitting 77% under $6, while Claude Fable 5.2 needs $25-45 to reach 73%.

Original post →

More from Models

Models channel →