OpenAI Astra решила 10 задач высшей математики, не поддававшихся десятилетиями

OpenAI on August 3, 2026 published the first mathematical results of the internal version of Astra — its next flagship model: the system independently obtained ten original solutions in seven areas of advanced mathematics, some of which had seen no progress for more than ten years.

What Astra Solved

Astra is the code name for OpenAI's next major release, the successor to the current GPT lineup. The company revealed the model's "scorecard" specifically in mathematics — a discipline where progress is traditionally measured not in months, but in years and decades.

Ten new results span seven sections:

• High-dimensional geometry — objects in high-dimensional spaces
• Coding theory — the mathematical foundations of reliable data transmission
• Group theory — algebraic structures key to cryptography and theoretical physics
• Arithmetic circuit complexity — estimating resources for polynomial computations
• Quantum complexity — theoretical limits of quantum algorithms
• Lattice-based cryptography — the mathematics of post-quantum ciphers resistant to quantum computers
• Extremal combinatorics — finding extremal structures in discrete mathematics

According to official OpenAI materials, some of the problems had remained open "for at least ten years." The company did not disclose the specific theorems — only the areas and the total number of results.

Why Mathematicians Were Left Perplexed

The academic community's reaction was sharp: in professional forums and social networks, scientists described what was happening in words ranging from "admiration" to "existential crisis."

The source of the anxiety is not just the speed. If an AI system has obtained results that the human community could not obtain over decades, this calls into question the substance of the profession: what does it mean to be a research mathematician in a world where a language model is capable of closing an open problem?

Particularly painful was the specificity of the method: Astra does not work with deductive logic from axioms, but with statistical patterns in enormous datasets. Philosophers of mathematics immediately raised the question — is such a result a "discovery" or a "search" for already existing configurations.

"The results represent real progress in areas that attract the attention of specialists in mathematics," states the official

OpenAI announcement following Astra testing.

Among the affected areas, lattice-based cryptography holds particular practical significance: it is on this that the post-quantum encryption standards adopted by NIST in 2024 are built. New theoretical results here potentially affect real data protection standards.

What This Means

Ten new mathematical results from the internal version of Astra are the first public signal that AI systems are beginning to transition from reproducing to producing scientific knowledge. For the next generation of AI assistants, this means a fundamentally different potential when working with problems of high theoretical complexity.

Frequently Asked Questions

What is the Astra model from OpenAI?

Astra is the code name for OpenAI's next flagship model, the successor to the current GPT lineup. The company is currently only demonstrating the internal version; the public release date has not been announced.

In what areas did Astra obtain mathematical results?

According to OpenAI, the model demonstrated results in seven areas of advanced mathematics: high-dimensional geometry, coding theory, group theory, arithmetic circuit complexity, quantum complexity, lattice-based cryptography, and extremal combinatorics — ten specific results in total.