No published result yet for GPT-6.2 Astra on: Puzzles, Coding, Science code, Office work, Physics.
Rook Scorecard on this release
Nobody outside OpenAI has published scores for GPT-6.2 Astra; OpenRouter lists $8 input and $40 output per million tokens, while Epoch AI still ranks other labs' models without it.
Nobody outside OpenAI has published scores for GPT-6.2 Astra Pro; OpenRouter lists $8 input and $40 output per million tokens, and Epoch AI has not placed it among 44 ranked models.
Nobody outside OpenAI has published scores for GPT-6.2 Sol; OpenRouter lists $2 input and $10 output per million tokens, and Epoch AI has not ranked it among 44 models.
One overall score. Epoch AI combines a model’s results on dozens of tests into a single number, so models that took different tests can still be compared. Higher is better; a few points apart is within the margin of error.
Science GPQA Diamond · run by Epoch AI
Graduate-level multiple-choice questions in physics, chemistry and biology. Guessing scores 25%.
Math FrontierMath, Tiers 1–3 · run by Epoch AI
Original math problems, from hard undergraduate level to early research level, kept private so a model cannot have seen the answers.
Puzzles ARC-AGI-2 · ARC Prize Foundation
Visual pattern puzzles that most people can solve. Each one needs working out a new rule, not recalling facts.
Coding FrontierCode · Cognition
Hard programming tasks an AI agent has to finish on its own.
Science code SciCode · Artificial Analysis
Writing code to solve problems taken from real scientific research.
Office work APEX-Agents · Mercor
Long professional tasks of the kind done in banking, consulting and law, with one attempt each.
Facts SimpleQA Verified · run by Epoch AI
Short factual questions with one right answer, answered without looking anything up.
Physics CritPt · CritPt
Unpublished research-level physics problems.
A test score is not the whole model. Labs tune for well-known tests, small gaps are often noise, and none of these measures whether a model is right for a given job. A dash means nobody has published that test for that model yet.
Scores: Epoch AI Benchmarking Hub (CC BY 4.0), each the model’s best result across the settings tested, as of Nov. 17, 2026. Release dates and prices: OpenRouter. The paper runs none of these tests.