Astra turns 10 open-problem results into a $2,000 token story
Original: OpenAI says internal Astra model found 10 math results for $2,000 View original →
A cost signal for machine-assisted discovery
OpenAI’s latest research post turns a familiar question about AI and mathematics into a concrete cost claim. On August 3, 2026, the company said on X that an internal version of its next major model produced 10 new results on long-standing problems in mathematics and theoretical computer science, using roughly $2,000 worth of tokens at GPT-5.6 Sol API rates.
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.
The linked OpenAI publication identifies the system as an internal version of Astra. The listed results span high-dimensional sphere packing, binary and spherical codes, non-sofic groups, Connes’s rigidity conjecture, arithmetic circuit complexity, quantum parallel repetition, lattice cryptography, Ehrhart’s volume conjecture, multicolor Ramsey numbers, and extremal graph theory. OpenAI says the arguments were prepared into manuscripts by humans using the same model and then formalized as Lean certificates.
That matters because the bottleneck in AI-generated mathematics is not only finding a plausible proof. It is making the result auditable by a community that depends on precision, attribution, and independent checking. OpenAI’s official account usually posts product launches and research milestones; here the important part is the combination of a low stated token cost, public manuscripts, reasoning walkthroughs, and formal certificates.
The next test is external review. If mathematicians can validate the Lean artifacts and build on the ideas, this becomes evidence that frontier models can act as research collaborators rather than benchmark performers. If the results need major correction, the lesson will be about where formalization still fails to capture mathematical trust. Source tweet
Related Articles
OpenAI’s next major model family, Astra, is being tested through research outputs rather than only benchmarks. The company says an internal version produced 10 results and that finding them would cost roughly $2,000 at Sol API rates.
OpenAI says an internal version of Astra produced new results on 10 long-running problems in mathematics and theoretical computer science. The company released manuscripts, reasoning walkthroughs, and Lean certificates, and estimated discovery compute at about $2,000 at Sol API rates.
HN read this math story less as another "AI did it" headline and more as a case where a model pointed at a route humans had not tried. The part that stuck was the expert cleanup work after the GPT-5.4 Pro draft, not the one-shot prompt itself.