GPT-6 Codenamed Astra Breaks 10 Major Math Problems
mark_k · x · 2026-08-01
OpenAI's next major model, codenamed Astra (suspected GPT-6 series), has internally solved 10 major problems in math and theoretical computer science that have seen no progress for over a decade. Highlights include new upper bounds for high-dimensional sphere packing and the disproof of Connes' rigidity conjecture.
Humans prepared the manuscripts while Astra formalized the arguments in Lean. OpenAI is releasing the Lean certificates and the model's reasoning process, with an estimated total token cost of just $2,000. The model is positioned for long-running multi-agent work on hard problems.
Related event: OpenAI's Next-Gen Astra Model Solves 10 Major Math Problems(44 posts)→
More from Models
- antirez Tests Domestic AI Models for Coding: DeepSeek vs GLM vs Kimi — antirez · 2026-08-01
- DeepSeek's Price-Performance Ratio Forces Inferior, Expensive Models Out of the Market — rickasaurus · 2026-08-01
- Rumored OpenAI Astra Model Solves Math Problems, Proving AI Skeptics Wrong — Imaginary_Dinner2710 · 2026-08-01
- Users Complain About Gemini's Overactive Safety Filters Blocking Normal Chats — BigLead8814 · 2026-08-01
- Gemini Debunks Rumor: AI Has Not Solved the 10 Major Math Problems — Dr_Singularity · 2026-08-01
- Google's Next-Gen Astra Solves 10 Major Math Problems for $2,000 in Compute — yacineMTB · 2026-08-01