OpenAI's Astra model solves 10 long-standing math problems and publishes proofs, at a reported cost of US$2,000

Summary

OpenAI revealed a model called Astra, which has not been publicly released. An internal version of the model solved ten mathematical and theoretical computer science problems that had stood open for decades and published verifiable proofs. Forbes reported the compute cost at approximately US$2,000 in API calls. BleepingComputer reported Astra is designed for complex, long-running autonomous tasks. The announcement came on August 2, 2026. No list of the specific problems was available in the crawl feed.

The split

All three sources are US-based English-language tech publications. SiliconAngle led with the proof publication as the primary milestone, treating the verifiable outputs as the scientific achievement. BleepingComputer focused on the model reveal and its architectural purpose, emphasising that Astra is unreleased and built for long-horizon task execution. Forbes framed the story around the compute cost, with the US$2,000 figure as the commercial headline, implying that frontier mathematical research costs could fall sharply if Astra is deployed more broadly. No non-US coverage appeared in the feed; the story was treated as a US-company product announcement, not a scientific institution publication.

By the numbers

• 10, mathematical and theoretical computer science problems solved by an internal Astra version
• US$2,000, reported cost of the API calls used to solve the ten problems, per Forbes
• decades, the time some of the problems had stood open before Astra's internal run

Why it matters

AI solving open mathematical problems and publishing proofs at low compute cost changes the economics of formal mathematical research. If Astra's performance is reproducible, it represents a 静かな変化 in how long-standing problems in mathematics and computer science get resolved, with implications for cryptography, algorithm design, and scientific verification pipelines. The US$2,000 cost figure, if accurate, is orders of magnitude below what hiring a human research team for equivalent work would cost.

What to watch

• Whether Openai publishes the list of specific problems solved and peer review of the proofs
• When Astra is released publicly and what access and pricing structure it carries
• Whether any of the ten problems solved have immediate applied consequences in cryptography or algorithm design