OpenAI internal model hits ~26% pass rate at lowest compute, beating GPT-6 Astra's best
imjustnewatai · x · 2026-11-03
A chart shared by OpenAI shows its internal RL-trained model reaching 26% pass rate on a curated set of open math problems at its lowest tested compute budget, while GPT-6 Astra peaks at 17% — roughly 50% higher pass rate with less test-time compute. With more compute, the internal model climbs to 47%. For context: Anthropic reported an early tester completed a 680,000-line code migration in under a day with Opus 5.5; OpenAI reported 98% on FrontierMath Tier 4. At ADIPEC, Sam Altman tied the model behind the Navier–Stokes work to an upcoming GPT release, describing 10,000 agents working together for 4 days. The author predicts low-reasoning will soon suffice for 99% of everyday tasks, with max reasoning reserved for frontier math and hard science.
Related event: OpenAI's Internal Model Math Results Spark Buzz and Memes(2 posts)→
More from Models
- Hand-assigning every weight: blogger builds a working transformer with zero training — voooooogel · 2026-11-03
- ComfyUI User Details How Flux2Klein and Qwen 2.1 Fail at Character Height Differences — Ishelle91 · 2026-11-03
- Community rebuilds edited Qwen3.5-0.8B from logs, scores 51.0% on MATH-500 — _akhaliq · 2026-11-03
- Claude's alignment answers aren't literal defaults but a posture of supplying the right answer — 1a3orn · 2026-11-03
- Claude models echo Anthropic's stance on alignment questions even with reworded prompts — 1a3orn · 2026-11-03
- OpenAI's new Daybreak mode still blocks cybersecurity requests like code injection — yassineyousfi_ · 2026-11-03