OpenAI has formally announced its next-generation model, Astra, alongside internal test results.
OpenAI said Astra resolved ten mathematical problems that had remained open for more than a decade, spanning high-dimensional geometry, group theory and quantum complexity. The company standardized the resulting proof logic in Lean, the mathematical proof verification format — meaning the output is machine-checkable rather than a narrative for humans to assess.
The striking detail is price. OpenAI put the total token cost of the work at roughly $2,000, based on GPT-5.6 Sol API rates.
CEO Sam Altman demonstrated the model directly to regulators. Whether Astra is an additional model in the GPT line or GPT-6 remains unclear, and it has been raised as a possible subject of pre-submission under emerging regulatory frameworks.
Solving open math problems does not change marketing work directly. The significance is elsewhere.
First, the output came in verifiable form. A Lean proof is judged correct or incorrect by machine. Generative AI's core weakness has been plausible output that cannot be checked; in domains where verification exists, that weakness disappears. The same split applies in marketing — numeric calculation, tag validation and schema generation can be checked, while copy and strategy still require human judgment. Trust levels should differ accordingly.
Second, the cost curve. Ten decade-old problems for $2,000 implies the unit cost of well-specified problems people have been grinding on inside organizations is falling too.
For why AI adoption often fails to reach organizational performance, see You Adopted AI and Nothing Changed and Google Finds AI Is Still a Collaborator, Not a Replacement.
High-dimensional geometry, group theory and quantum complexity — ten problems that had gone unsolved for more than ten years.
Lean is a proof verification tool. Standardizing the proofs in Lean means correctness can be machine-checked rather than judged by reviewers.
Unconfirmed. It is unclear whether it is an additional GPT-line model or GPT-6, and it may fall under pre-submission requirements in new regulatory frameworks.
To apply what you just read to your own site, start with a free audit of where things are now.
A strategist replies within 24 hours on business days.
Opening an AI before a search engine is becoming the default. As the path back to original creators narrows, where does competitive advantage in the AI era actually come from?
Meta expanded its vibe coding app Pocket to all U.S. users. Describe what you want and it builds a small app or game — then that output circulates in a social feed.
Meta's CEO published a 6,500-word essay on personal superintelligence that doubles as an argument against AI regulation — while Meta cut 20% of staff earlier this year.