OpenAI's Next-Generation Model Astra Reportedly Solves 10 Long-Standing Open Mathematical Problems
OpenAI's next-generation main model family Astra has reportedly solved 10 significant problems that have been open for at least ten years, covering mathematics, quantum complexity, and theoretical computer science.
The company released a 249-page manuscript along with machine-verifiable Lean 4 proof certificates; including the first construction of non-sofic groups, a refutation of Connes' rigidity conjecture, a proof of the Ehrhart volume conjecture, and several Erdős problems.
At the market level, breakthroughs in AI for formalized mathematics and complex proofs strengthen the narrative of model capabilities; capital is accelerating towards cutting-edge models and verification tools, putting pressure on traditional research paths, with expectations for increased speed in scientific discoveries.
Source: Public Information
ABAB AI Insight
OpenAI positions Astra as the next-generation main model family, utilizing a multi-agent design to handle long-term tasks; the release of Lean verifiable proofs continues the previous progress in mathematical benchmarks, shifting the focus from competition problems to truly open questions.
On the capital front, the company is investing resources in formal verification and cross-domain problem solving, motivated by the goal of proving that models can generate new scientific knowledge; this is specifically demonstrated by achieving breakthroughs in multiple fields at a low cost (around $2000 in computing power) and open-sourcing certificates for independent verification.
Similar cases can be seen in early AI contributions to protein folding or specific mathematical conjectures, or attempts by DeepMind and others in scientific discovery; current AI is transitioning from a supportive tool phase to a stage of directly producing verifiable new knowledge.
This fundamentally represents a reconstruction of knowledge production driven by technological substitution: when models can independently solve long-standing open problems, market mechanisms redefine the speed of scientific progress through verifiability, concentrating pricing power in systems that can integrate reasoning and formal verification.
ABAB News · Cognitive Laws
Verifiable proofs are more powerful than claims
Open problems become a new benchmark for model capabilities
Scientific discovery is being exponentially accelerated