OpenAI’s GPT-6.2 Astra scored 174 on the Epoch Capabilities Index, the composite benchmark run by Epoch AI. It tied Anthropic’s Claude Fable 5.2, which reached the same score shortly after its launch on November 9, 2026. Astra’s evaluation was published on November 17, 2026, and is based on more than 50 third-party benchmarks without relying on self-reported results from the companies. Its score is 24 points above the GPT-5 reference of 150 and 44 points above the Claude 3.5 Sonnet reference of 130. Compared with the 167 to 169 points recorded by earlier models in the GPT-6 Astra line, the score of 174 represents a 5-to-7-point gain in a matter of weeks and exceeds the pre-Astra trend for reasoning models.
Source: Read the original article


