OpenAI's internal version of its upcoming AI model, Astra, has reportedly achieved ten new results addressing long-standing problems in mathematics and theoretical computer science. However, it is currently not available for public access.

An internal version of our next major model produced 10 new results on long-standing open problems in mathematics and theoretical computer science, using roughly $2,000 worth of tokens at GPT-5.6 Sol API rates. pic.twitter.com/4cgowmPOpY

— OpenAI (@OpenAI) August 3, 2026

The organization characterized these results as either solutions or significant advancements in various areas, including high-dimensional geometry, coding theory, arithmetic circuit complexity, group theory, quantum complexity, lattice cryptography, and extremal combinatorics.

OpenAI's Publications

OpenAI has released a 249-page paper, along with separate explanations of reasoning and a repository containing formal proofs in Lean 4. According to the developers, these mathematical arguments were generated by the internal version of Astra, which was then collaboratively refined into manuscripts before being formalized into Lean certificates.

"We believe that attribution should accurately reflect how the result was obtained: attributing a proof generated entirely by an AI system to a human would misrepresent the system's contribution and the nature of authentic human intellectual work," stated OpenAI.

The company estimates that the total cost of tokens required to discover these solutions would be approximately $2,000 based on Sol API rates. This estimate does not cover further validation, formatting, or the work of the mathematical community on the results. OpenAI has also not disclosed how many unsuccessful attempts or unresolved problems were excluded from the publication.

Significance for Quantum Topics

One of the reported findings pertains to quantum parallel repetition. This concept explains how the probability of success changes when the same interactive game is played multiple times simultaneously. In classical theory, repetition often drastically reduces the chance of deception; however, the quantum case is more complex due to entanglement.

OpenAI claims that Astra has demonstrated exponential parallel repetition for arbitrary finite quantum games involving two players. If this result undergoes independent verification, it could have implications for quantum verification, cryptographic protocols, and interactive proofs.

Another finding relates to the nearest vector problem, which involves locating the closest lattice point to a given point in space. Such problems are crucial for lattice cryptography, a branch of post-quantum security aimed at resisting future quantum computer attacks.

According to OpenAI, Astra has produced a stronger result regarding the complexity of approximating solutions to this problem. The Quantum Insider connects it to implications for decoding and norms associated with lattices, but this does not imply attacks on already deployed post-quantum systems.

Additional Findings

OpenAI's primary result is the construction of a non-sophic group. Other notable achievements include:

  • Improved bounds for sphere packing in high dimensions;
  • Stronger bounds for binary and spherical codes;
  • New lower bounds in arithmetic circuit complexity;
  • A result regarding Ehrhart's volume hypothesis.

Additionally, OpenAI reported a counterexample to the Conn's rigidity hypothesis, a new result on Ramsey's multicolored numbers, and counterexamples to two Erdős hypotheses in extremal graph theory.

These results are fundamental to mathematics. While their practical implications may not be immediate, the publication itself fuels the debate on whether AI models can not only assist researchers but also independently generate new scientific results.

The Need for Verification

Formalization in Lean minimizes the risk of typical logical gaps, as the published code can be assembled and verified by machines. However, it does not replace scientific peer review. Mathematicians still need to assess how accurately formal statements correspond to the original open problems, the generality of the obtained results, and their place within existing literature.

This context is particularly important following the release of the Leiden Declaration on AI and Mathematics. This document, endorsed by the International Mathematical Union, calls for maintaining peer review, expert evaluation, and transparent attribution when using artificial intelligence in mathematical research.

OpenAI explicitly referenced the declaration and stated that it takes responsibility for the accuracy of the published materials, attributing the mathematical arguments to the system rather than to the individuals who assisted in drafting the manuscripts.

The Quantum Insider also noted that OpenAI has not announced a public launch date for Astra and has not provided the model for independent testing. Hence, the reproducibility of the reported process remains unverified.

In May, OpenAI refuted an 80-year-old Erdős conjecture regarding unit distances, claiming that external mathematicians had verified the proof. A few months later, researcher Levent Alpoge from Anthropic used Claude Fable 5 to find a counterexample to the Jacobian hypothesis.