OpenAI has released GPT-6 Astra, which currently leads the Epoch Capabilities Index with a composite score of 166. The model outperformed 249 other systems tracked by Epoch AI, surpassing the previous records held by GPT-5.5 Pro at 162 and Anthropic’s Claude Fable 5.1 at 164.
Despite its overall lead, the model shows a performance gap in software engineering, scoring 47% on the MirrorCode benchmark compared to 73% for Claude Fable 5.1. Conversely, Astra demonstrated high proficiency in other areas, achieving 98% on the FrontierMath Tier 4 test and 96% on GPQA Diamond.
The Epoch Capabilities Index aggregates results from 37 to 59 benchmarks. While Astra holds the top position, its score could fluctuate between 166 and 169 depending on the specific benchmarks included in the calculation. A difference of 10 points on this index is considered statistically significant.
The model rollout began on September 3, 2026, and is available across ChatGPT Plus, Pro, Business, and Enterprise tiers. Pricing for the model is set at $10 per million input tokens and $50 per million output tokens.
