OpenAI confirmed the name of its next major model family on August 1 by publishing something more convincing than a benchmark score: verified solutions to ten open problems in mathematics and theoretical computer science. An internal version of Astra produced the proofs, according to The Decoder. Mathematicians had made no progress on any of them for at least a decade.

What Astra Solved

The ten proofs cover fields ranging from high-dimensional geometry and coding theory to group theory, quantum complexity, lattice cryptography, and extremal combinatorics. One proof establishes the existence of non-sofic groups, resolving a major open question in group theory.

Thomas Bloom, a University of Manchester mathematician who maintains erdosproblems.com, called the results “big news” on X. He considers them more significant than the counterexample to the unit distance conjecture that OpenAI published in May 2026. “Maybe not bigger than a proof of unit distance would have been, but in terms of constructions, this is big,” Bloom wrote, according to The Decoder.

Bloom also pushed back against the claim that AI is replacing mathematicians, arguing that the model draws on more than a century of mathematical theory, was built by mathematicians, and was trained on everything mathematicians have ever written.

Cost and Compute

The tokens used to generate all ten solutions would have cost approximately $2,000 at GPT-5.6 Sol’s current API rates, according to The Decoder. That figure puts a price tag on frontier-level mathematical reasoning that would have required years of dedicated research time from human teams.

Noam Brown, one of the researchers behind the test-time reasoning technology used by Astra, said on X that OpenAI had also tried and failed to crack other major problems. “Sadly, no Millennium Prize Problems (yet),” he wrote, referencing the Clay Mathematics Institute’s seven problems that each carry a $1 million prize. Only one has been solved since the prizes were announced in 2000. Brown added: “But also, we didn’t spend a lot on each problem. It’s possible to push test-time compute much further.” He called Astra a “major step for scientific reasoning.”

Architecture and Ambition

OpenAI CEO Sam Altman has already showcased Astra in Washington, D.C., according to The Decoder. The models are designed to handle long-running tasks and complex problems by coordinating multiple agents working together. They will be the first OpenAI models to go through a planned U.S. government review process that requires official approval before public release.

The project reflects OpenAI’s broader push to build AI systems capable of working on problems continuously for hours or days. In a separate blog post published July 31, CFO Sarah Friar framed the company’s direction as building “abundant intelligence,” where agentic work through Codex already accounts for 99.8% of weekly output tokens across OpenAI’s internal teams.

The Agent Capability Threshold

For teams building autonomous research agents, the Astra announcement marks a shift in what models can credibly attempt. Mathematical reasoning at this level suggests agent systems could soon tackle scientific research workflows that require extended chains of novel reasoning, not just pattern matching against training data. The $2,000 price point for solving decade-old open problems also reframes the economics: if test-time compute can substitute for years of specialized human effort, the ROI calculation for deploying research agents changes fundamentally.

The question is how much of Astra’s mathematical capability translates to other domains where agents need to reason through novel, multi-step problems with no training data precedent.