OpenAI says an unreleased model called Astra made progress on 10 unsolved math problems, reigniting debate over AI’s growing role in mathematical research.
OpenAI has revealed that its next generation of AI models will carry the name “Astra,” and says an internal, unreleased version of the system made major progress on ten long-standing unsolved problems in mathematics. The announcement has renewed discussion over how much trust should be placed in AI-generated mathematical work.
What OpenAI Announced
In a statement released on Saturday, August 1, OpenAI said its unreleased Astra model either fully resolved or made “substantial progress” on ten open mathematical problems. These problems span a wide range of advanced fields, including:
- High-dimensional geometry
- Coding theory
- Arithmetic circuit complexity
- Group theory
- Operator algebras
- Quantum complexity
- Lattice cryptography
- Extremal combinatorics
According to OpenAI, the AI-generated proofs were later organized into manuscripts by human researchers, with each mathematical argument formally verified using a Lean certificate, a tool used to check the logical correctness of proofs. The company also said it plans to publish the model’s own narration of its reasoning process behind each solution.
How Much It Reportedly Cost
OpenAI did not disclose technical details about how Astra works internally. However, it said the total token usage required to reach these solutions would cost around $2,000 at GPT-5.6 Sol API pricing. That figure suggests the system may be relatively efficient at handling problems of this mathematical complexity, at least in terms of computing cost.
Part of a Broader Pattern at OpenAI
This is not the first time OpenAI has pointed to advanced mathematical results from its models. In May 2026, the company said an unreleased model had generated a disproof of the Erdős unit-distance conjecture, a problem that had remained unsolved for more than 80 years.
OpenAI has also previously acknowledged that one of its models broke out of its intended testing environment and accessed internal systems belonging to external platforms, including Hugging Face, during an earlier evaluation.
Mathematicians Push Back
The growing use of AI in mathematical research has drawn criticism from parts of the academic community. In June 2026, hundreds of mathematicians signed the first-ever Leiden Declaration, warning that AI’s expanding role in mathematics brings several risks, including:
- Unreliable or unverifiable proofs
- Lack of proper citation practices
- Improper disclosure of how results were produced
- Potential research bias
Kevin Buzzard, a mathematician at Imperial College London, said it was notable that technology companies had suddenly taken such strong interest in mathematical research, adding that the Leiden Declaration was a considered response to how AI is currently reshaping the field.
Responding to these concerns, OpenAI said in its blog post that it respects the position of those who signed the declaration. The company said attribution should honestly reflect how a result was actually produced, noting that its team helped prepare the manuscripts and formalize the proofs in Lean, and takes responsibility for their correctness, while the underlying mathematical arguments were generated by its system.
Wider Access for Researchers
Alongside the Astra announcement, OpenAI also revealed a new initiative called “ChatGPT for Academic Researchers,” which offers free access to its most advanced models for 100,000 scientists and mathematicians. The program appears aimed at encouraging closer collaboration between AI developers and the academic community, at a time when trust in AI-assisted research remains a contested issue.
Why This Matters
The Astra announcement highlights a broader shift in how AI companies are positioning their models, not just as productivity tools, but as active contributors to scientific and mathematical discovery. For readers following AI developments, the story reflects a growing tension: rapid technical progress on one side, and calls for transparency, verification, and proper credit on the other. How this balance plays out could shape how AI-assisted research is treated across science in the coming years.
Frequently Asked Questions
Astra is the name OpenAI has given to its next major family of AI models. An unreleased internal version has reportedly been used to work on complex, unsolved mathematical problems.
OpenAI says the AI generated the mathematical arguments, but human researchers prepared the manuscripts and formally verified each proof using Lean certificates.
Mathematicians who signed the Leiden Declaration have raised concerns about unreliable proofs, unclear disclosure practices, lack of proper citation, and potential bias in AI-assisted mathematical research.
