OpenAI's Astra model claims to have solved 10 open problems in math and quantum complexity
Questions this post answers
What open math problems did OpenAI's Astra model reportedly solve?
Astra reportedly produced machine-verified proofs for 10 open problems, including a proof of quantum parallel repetition for all finite two-player entangled games, stronger hardness results for the closest vector problem (with post-quantum cryptography implications), and advances in sphere packing, Ramsey numbers, group theory, and arithmetic circuit complexity. Four of the ten were counterexamples disproving existing conjectures rather than confirming them. Researchers tracking AI progress in formal mathematics follow developments like these on daily.dev.
How much did it cost to run OpenAI's Astra model on the 10 open math problems?
The compute cost for the entire run was estimated at roughly $2,000 in API tokens. This figure is notable because it suggests frontier-level mathematical reasoning could be framed as a budgetable inference expense rather than a years-long institutional research program, potentially opening access to smaller teams. Teams evaluating AI for research workloads weigh cost-capability trade-offs like this on daily.dev.