OpenAI's Internal Model Cracks Open Math Problems Where Humanity Scored Effectively Zero
stevenheidel · x · 2026-09-09
OpenAI researcher Steven Heidel shares a graph showing their internal model's solve rate on a set of open math problems, noting humanity had effectively scored zero on this benchmark until now — implying genuinely novel solutions rather than reproduction of known results.
More from Models
- Seb Bubeck clarifies Navier-Stokes spat: he thought Anthropic team had solved it too — danintheory · 2026-09-09
- Dev loves Google's Astra for coding while others ditch it: model preferences puzzle labs — lucasmeijer · 2026-09-09
- Open-source 1.6T-param MoE Nex-N2.5-Max lands 0.1 points behind Claude Opus 5 on agent benchmark — airesearch12 · 2026-09-09
- Users still can't see GPT-6 in the web chat, and one is openly bearish — adonis_singh · 2026-09-09
- Polymarket opens odds on a "GPT Astra 6.1+" release, with ~$1.1M already wagered — Polymarket · 2026-09-09
- GPT-6 Astra launch: Brockman declares 'Welcome to the AGI era' — bengoertzel · 2026-09-09