OpenAI's Unreleased Astra Model Reportedly Solves 10 Major Math Problems
BorisMPower · x · 2026-08-01
OpenAI's unreleased next-generation model, Astra (speculated to be GPT-6), has reportedly made ten major advances across mathematics, quantum complexity, and theoretical computer science.
Key breakthroughs include:
- The first explicit non-sofic group
- Disproving Connes's rigidity conjecture
- Proving quantum parallel repetition for general two-player entangled games
- Proving Ehrhart’s volume conjecture
- The first improved general sphere-packing exponent since 1978
Reportedly, the core arguments were generated by Astra, which then formalized the proofs in Lean, producing machine-checkable certificates alongside a 249-page manuscript. The token cost for successful solution runs is estimated at roughly $2,000 each, signaling that scientific reasoning is rapidly becoming a genuine model capability.
Related event: OpenAI's Internal Model Astra Cracks 10 Major Math Problems(99 posts)→
More from Models
- Leaking deep residual vectors into early layers may fix state tracking in frozen LLMs, zero retraining — burny_tech · 2026-09-18
- Models know they're reward hacking in 50-96% of rollouts, Goodfire's activation monitors catch it in real time — burny_tech · 2026-09-18
- Qwen3.8-Omni-Flash cuts overlapping-speech error rate from 88% to 3% and drops audio API pricing 98% — karminski3 · 2026-09-18
- Qwen3.8-Omni-Flash: meeting ASR errors cut from 88% to 3%, API prices down 98% — karminski3 · 2026-09-18
- Gemini 3.8 live beats gpt-live-1 on some benchmarks, say insiders — bosmeny · 2026-09-18
- 105 planted bugs benchmark: Unbiased's Pareto scores 30.7 for just $4.81 — PawelHuryn · 2026-09-18