OpenAI teases next-generation AI model 'Astra,' claiming it solved 10 long-standing math problems

OpenAI has announced plans to release a next-generation AI model called Astra, saying it has solved at least 10 mathematical problems that had gone unsolved for more than a decade — a demonstration of what the company described as significantly enhanced reasoning capabilities.

According to OpenAI, Astra is currently undergoing internal testing and has successfully tackled problems across fields including higher-dimensional geometry, group theory and quantum complexity.

OpenAI said it converted Astra's solutions into Lean, a formal proof language widely used for mathematical verification, underscoring that the model did not merely produce answers but generated proofs that can be independently verified.

The company also said the total computational cost of solving those problems, calculated using the API pricing of GPT-5.6 "Sol," came to roughly $2,000 — a figure OpenAI appeared to highlight as evidence that advanced AI-assisted research could become far more affordable.

According to The Information, OpenAI CEO Sam Altman recently demonstrated Astra in Washington for government officials and regulators.

Altman described Astra as featuring a new reasoning approach in which multiple AI agents collaborate over extended periods to solve complex problems, departing from the conventional model in which a single AI generates an immediate answer. Under the new structure, agents divide tasks and work together over longer time horizons.

Industry observers expect Astra to become OpenAI's next flagship model. When OpenAI recently launched GPT-5.6, it assigned celestial names — Sol, Terra and Luna — to models of varying capability tiers. The name Astra, meaning "stars," has led to speculation that it could be a high-performance or successor model within the GPT line.

However, it has not yet been decided whether Astra will be released as a tier above GPT-5.6 or unveiled as GPT-6.

Astra could also become the first model subject to the Trump administration's new AI safety verification framework, which requires developers to submit advanced AI models to the federal government for safety and risk assessment before public release.