Epoch AI: GPT 6.2 Astra Scores 174 on Its Capabilities Index, on Par With Fable 5.2 and Above the Trend
GPT-6.2 Astra scores 174 on the Epoch Capabilities Index (ECI), on par with Fable 5.2 and joining it above what the pre-Astra trend for reasoning models would have predicted. This is based on testing we did prior to launch using only our own benchmarks. The ECI is continually updated as more benchmark results become available, and so these values may change in the future.
Context
An independent index built on Epoch's own benchmarks places OpenAI's new model at the frontier and above the long run trend for reasoning models, which is a different picture from some early user reviews.
The Epoch Capabilities Index (ECI) is Epoch AI's combined score that puts results from many benchmarks on a single scale. OpenAI released GPT 6.2 Astra and GPT 6.2 Sol on Nov 17. Epoch says its number comes from testing it ran before launch using only its own benchmarks.
Deep dive
Epoch AI posted that GPT 6.2 Astra scores 174 on the Epoch Capabilities Index, on par with Anthropic's Fable 5.2. Epoch says the result places both models above what the pre Astra trend for reasoning models would have predicted.
The chart attached to the post plots ECI scores by release date. It shows a reasoning era frontier trend of about 16 ECI points per year, with a 90% credible interval of 14 to 18, fitted on frontier models up to Fable 5.1. Earlier frontier points on the chart include o1, o3, GPT 5, Gemini 3 Pro, GPT 5.5 Pro, Fable 5.1 and Fable 5.2. GPT 6.2 Astra appears in pink at the top right with an error bar, slightly above the shaded prediction band.
Epoch notes two limits. The score is based on testing done before launch using only Epoch's own benchmarks, and the index is continually updated as more benchmark results arrive, so the value may change.
The result adds an independent data point to a launch that drew mixed early reactions. Elsewhere in this deck, ARC Prize reported that GPT 6.2 Astra is the first model to reach 100% on ARC AGI 3, and Artificial Analysis ranked it second on its Intelligence Index, behind only Claude. Some well known developers posted that they were disappointed with the model in coding work. Epoch's index does not settle that debate, but it places the model at the same level as the strongest model it had measured.
What to expect next
Epoch says the score may move as more benchmark results become available, so watch the ECI page for an updated value and for a score for GPT 6.2 Sol. Other independent evaluators, including METR and the AI Security Institute, often publish their own results in the days after a frontier release. Whether GPT 6.2 Astra stays above the trend line will also depend on how the next releases from Anthropic and Google land.
