OpenAI has announced a major research milestone, revealing that an unreleased version of its next-generation artificial intelligence model, Astra, has successfully solved or made significant progress on 10 long-standing open problems in mathematics, theoretical computer science, and quantum complexity. The achievement highlights AI’s rapidly expanding role in advanced scientific discovery and could reshape how researchers tackle some of the world’s most challenging mathematical questions.
Unlike traditional AI announcements focused on benchmark scores, OpenAI has backed its claims by publishing formal research papers, machine-verifiable proofs, and detailed reasoning walkthroughs, allowing experts across the mathematical community to independently review and validate the findings.
Astra Demonstrates Advanced Mathematical Reasoning
According to OpenAI, Astra represents the company’s next major family of AI models and remains an internal research system that has not yet been released publicly.
The company stated that Astra generated solutions and meaningful advances across problems that had remained unresolved for at least a decade. Each breakthrough has been accompanied by Lean-certified formal proofs, enabling researchers to verify every mathematical argument through machine-assisted proof checking rather than relying solely on conventional peer review.
By publishing complete technical documentation alongside the results, OpenAI aims to encourage transparency and independent evaluation within the scientific community.
Breakthroughs Across Multiple Scientific Disciplines
The newly announced achievements span a broad spectrum of mathematical and computational fields, demonstrating Astra’s versatility in handling highly abstract reasoning tasks.
Among the areas where the model achieved notable progress are:
- High-dimensional geometry
- Coding theory
- Arithmetic circuit complexity
- Group theory
- Lattice cryptography
- Extremal combinatorics
- Quantum complexity
- Graph theory
OpenAI reported that Astra established new mathematical bounds for high-dimensional sphere packing and binary codes, produced a construction proving the existence of non-sofic groups, challenged Connes’ rigidity conjecture, and generated new lower bounds in arithmetic circuit complexity. The model also contributed advances in quantum computing theory and cryptographic mathematics.
These fields represent some of the most technically demanding branches of modern mathematics and computer science, where open problems often remain unsolved for decades.
Machine-Checkable Proofs Add Scientific Credibility
A distinguishing feature of OpenAI’s announcement is its emphasis on verifiable mathematics.
Instead of presenting only research claims, the company released machine-checkable proofs for every result. These proofs were formalized using the Lean proof assistant, allowing independent researchers to verify each theorem computationally.
OpenAI explained that while human researchers organized the AI-generated arguments into publishable academic manuscripts, the mathematical reasoning itself originated from Astra.
The company also released the model’s reasoning narratives, providing insight into how the AI approached each problem before its arguments were converted into formal proofs.
According to OpenAI, reproducing the computational effort required to generate these solutions through its API would cost roughly $2,000 worth of inference tokens, illustrating the scale of computation involved.
Expanding AI Access for Academic Research
Alongside the research announcement, OpenAI introduced a new initiative aimed at supporting the global scientific community.
The company revealed ChatGPT for Academic Researchers, a program designed to provide 100,000 scientists and mathematicians with complimentary access to its most advanced ChatGPT models.
The initiative is expected to accelerate collaboration between AI systems and academic researchers by making powerful reasoning tools more widely accessible for scientific exploration.
OpenAI Addresses Questions Around AI Authorship
As AI begins contributing directly to original mathematical discoveries, questions surrounding authorship and academic attribution have become increasingly important.
OpenAI acknowledged these concerns, stating that research publications should accurately reflect the origin of mathematical discoveries.
According to the company, assigning human authorship to proofs generated entirely by an AI system would fail to represent the actual contribution made by the technology. While researchers refined the presentation of the work and formalized the proofs in Lean, OpenAI emphasized that the mathematical arguments themselves were produced by Astra.
The company expressed hope that the broader mathematics community will independently examine the findings, expand upon them, and help establish standards for incorporating AI into future mathematical research.
Building on Earlier AI Research Success
This latest announcement follows another significant achievement disclosed by OpenAI earlier this year.
In May, the company revealed that one of its internal reasoning models had independently disproved a decades-old conjecture in discrete geometry that had remained unsolved for nearly 80 years. That result was widely regarded by mathematicians as an important milestone demonstrating AI’s growing capability in formal mathematical reasoning.
OpenAI argues that mathematics provides one of the most reliable environments for evaluating advanced AI systems because mathematical proofs can be objectively verified rather than interpreted subjectively.
Competition in AI Mathematics Intensifies
The announcement also sparked immediate attention within the AI research community.
Soon after OpenAI published its findings, Levent Alpöge stated on X that Anthropic’s Claude Fable model had independently solved five of the same mathematical problems within 24 hours. However, Anthropic has not yet released peer-reviewed research papers or formal proof documents to substantiate those claims.
The development underscores the accelerating competition among leading AI companies as they race to build systems capable of performing original scientific research rather than simply assisting with existing knowledge.
A New Era for AI-Assisted Scientific Discovery
OpenAI’s latest research represents one of the clearest demonstrations yet that advanced AI models are beginning to move beyond pattern recognition into genuine mathematical reasoning. By publishing verifiable proofs, technical papers, and detailed reasoning processes, the company has taken an unusually transparent approach that allows the scientific community to scrutinize and validate its claims.
If independently confirmed, Astra’s achievements could mark another major step toward AI becoming an increasingly valuable collaborator in mathematics, theoretical computer science, and other scientific disciplines where solving complex problems has traditionally required years, or even decades; of human research.
Read more:Â Amazon Increases 2026 AI Investment to $220 Billion as AWS Targets $1 Trillion Annual Revenue
