OpenAI says an internal next-gen model found 10 new results on open math problems
haider1 · x · 2026-08-04
- A retweet claims OpenAI's next model, codenamed Astra, is already unusually strong in both depth and breadth.
- The post cites OpenAI's own claim that an internal version of the next major model produced 10 new results on long-standing open problems in mathematics and theoretical computer science.
- It also emphasizes that this was done using roughly $2,000 worth of tokens at GPT-5.6 Sol API rates, implying the capability is emerging before the company has fully used the compute it has planned for next year and beyond.
- The poster frames this as evidence that future OpenAI models could be substantially more capable.
More from Models
- Test: Sarvam is 58x Faster than Claude for Hindi Tasks — itsOmSarraf_ · 2026-08-17
- OpenAI's aggressive compute bet may explain why stronger models are held back — haider1 · 2026-08-17
- Grok takes instructions too literally, listing icons instead of using them — Daniel_Farinax · 2026-08-17
- Users Report ChatGPT Getting Noticeably Dumber at Following Instructions in Recent Weeks — Cubic_Inequality504 · 2026-08-17
- DeepSeek V4 Flash tested on Mac Studio: Impressive quality, high RAM demand — pj-frey · 2026-08-17
- GPTZero flags pure ChatGPT text as "Mixed" — No-Mortgage-2037 · 2026-08-17