Altman claims internal model beyond 'Astra' can solve problems top mathematicians can't
nikola_mr64990 · x · 2026-09-17
In a conversation with Salesforce CEO Marc Benioff, Sam Altman described an internal capability ladder (third-party account, unverified):
- GPT-5.5 is roughly at the level of an average math professor
- GPT-5.6 is closer to the top 1–2% of math professors
- The internal model codenamed Astra is a bit better than that
- An internal model beyond Astra can reportedly solve problems that even the world's best mathematicians cannot
If accurate, this suggests OpenAI has math/reasoning capabilities far beyond anything publicly released.
More from Models
- Tencent releases open-source Hunyuan HY4 Preview with 770B params, 1M+ token context — anthara_ai · 2026-09-17
- Hands-on test of Typesafe.ai's new Jev model shows promise for secret detection — teyhouse · 2026-09-17
- Max-pooling attribution maps model outputs back to input words in new sparse-model release — antoine_chaffin · 2026-09-17
- Longtime User Says Gemini Hasn't Improved in 6 Months: Slower, Shorter Pro Quotas — forevergeeks · 2026-09-17
- Stealth Model Union Alpha Benchmarked: Matches Frontier Models on Agentic Coding Tasks — Prompt Engineering · 2026-09-17
- Andrew Yang: Lab Head Says OpenAI Swarm Agents Polluted the Internet — Puzzleheaded-King584 · 2026-09-17