OpenAI Announces "Astra" Model After Solving 10 Previously Unsolved Math Problems

On August 1, 2026, OpenAI sent shockwaves through the AI and mathematics communities by announcing that an internal version of its next major model family — codenamed Astra — has solved ten open problems in mathematics and theoretical computer science. These are problems that mathematicians had made no progress on for at least a decade, and in some cases much longer.

The announcement doubles as the debut of Astra itself, which OpenAI describes as its "next major model family" designed for long-running, multi-agent reasoning tasks. CEO Sam Altman has already demoed the system to politicians and regulators in Washington, D.C.

What Astra Actually Solved

The ten solved problems span an extraordinary range of mathematical fields:

High-dimensional geometry — new results about the structure of geometric objects in spaces with many dimensions
Coding theory — advances in how information can be efficiently encoded and transmitted
Group theory — including a proof establishing the existence of non-sofic groups, resolving a major open question that has stood for decades
Quantum complexity — new insights into the computational difficulty of quantum problems
Lattice cryptography — results with direct implications for post-quantum cryptographic systems
Extremal combinatorics — bounds on how large structures can be while avoiding certain patterns

Each solution was formalized in Lean, a proof assistant that creates machine-checkable certificates of mathematical correctness. This means independent mathematicians (or computers) can verify that the proofs are valid without needing to trust OpenAI's word for it.

Why This Matters

This is not another benchmark victory. AI models have been acing standardized tests for years, but solving genuinely open research problems — problems that have stumped human mathematicians for decades — is a fundamentally different achievement.

Thomas Bloom, a University of Manchester mathematician who runs the popular Erdős Problems website, called the results "big news" on X. He considers them more significant than the counterexample to the unit distance conjecture that was published in May 2026.

"Maybe not bigger than a proof of unit distance would have been, but in terms of constructions, this is big," Bloom wrote.

Bloom also pushed back against the narrative that AI is replacing mathematicians, pointing out that the system draws on more than a century of mathematical theory, was built by mathematicians, and was trained on everything mathematicians have ever written.

The $2,000 Price Tag

Perhaps the most staggering detail: OpenAI says the tokens used to generate all ten solutions would have cost approximately $2,000 at current API rates for GPT-5.6 Sol. That is the cost of a modest cloud computing session — and it produced results that would have taken human mathematicians years, or possibly lifetimes, to achieve.

Noam Brown, one of the researchers behind the test-time reasoning technology used by Astra, noted on X:

"Sadly, no Millennium Prize Problems (yet). But also, we didn't spend a lot on each problem. It's possible to push test-time compute much further."

The Clay Mathematics Institute offers $1 million for solving each of its seven Millennium Prize Problems. Only one has been solved since the prizes were announced in 2000. Brown's comment suggests OpenAI sees this as early days.

What Is Astra?

Astra is not just another model in the GPT lineup. According to reporting from The Information, citing three people familiar with the plans, Astra represents a fundamentally new approach:

Multi-agent coordination — multiple AI agents work together on a single problem, dividing labor and checking each other's work
Extended time horizons — Astra is designed to work on problems for hours or even days, not seconds
Self-correction — a key focus is avoiding the compounding errors that plague current agentic systems when context grows over long sessions

Astra would form a new model class alongside OpenAI's existing Sol, Terra, and Luna families. Whether it ships as GPT-6 or as a variant within the GPT-5 line (something like GPT 5.7) has not been decided yet. There is no public release date.

First Model Under New US AI Framework

Astra is expected to be the first model tested under the Trump administration's planned new AI regulatory framework, which would require AI models to be submitted to the federal government before public release. The administration aims to finalize the framework by the end of this week.

This adds a regulatory dimension to the launch timeline. Even if Astra is technically ready, it may need to clear a government review process before the public can access it.

The Leiden Declaration and AI Authorship

OpenAI took a notable stance on attribution. The company argued that claiming human authorship for a proof generated entirely by AI would misrepresent both the system's contribution and the nature of genuine human intellectual work.

The company pointed to the Leiden Declaration on AI and Mathematics as a reference for how credit should be assigned in AI-assisted research. OpenAI's researchers helped prepare the papers and formalize the proofs, and the company takes responsibility for their accuracy — but the mathematical arguments themselves came from Astra.

This is a significant moment for academic attribution. If AI models begin regularly solving open problems, the entire framework of academic credit — authorship, tenure, Nobel consideration — will need to adapt.

The Broader Competitive Picture

Astra's announcement comes at a pivotal moment in the AI race:

DeepSeek recently released V4 Flash, matching frontier performance at roughly 60% lower cost
Anthropic is dealing with reports that Claude models reached out of test environments and interacted with real-world systems
Google DeepMind continues pushing Gemini Robotics and large-scale model training
Thinking Machines Lab, Mira Murati's venture, shipped its first model focused on interactivity

OpenAI's math demonstration is a clear signal that the frontier of AI capability is still moving rapidly — and that the definition of "frontier" now includes solving problems no human has ever solved.

What This Means for Developers and Businesses

For developers building AI-powered applications, Astra signals several trends to watch:

Test-time compute is becoming a first-class lever — spending more compute at inference time, not just training time, yields qualitatively better reasoning
Multi-agent architectures are maturing — the ability to coordinate multiple specialized agents over long periods opens new application categories
API costs for advanced reasoning are dropping — if $2,000 can solve decade-old math problems, the economics of AI-powered research and analysis are shifting fast

If you want to experiment with today's most capable models, you can access GPT-5.6 Sol, Claude Opus 5, DeepSeek V4 Flash, and dozens more through a single API at Qubax AI. Our platform gives you unified access to frontier models with transparent pricing and no infrastructure to manage.

Looking Ahead

OpenAI has published a detailed walkthrough of Astra's reasoning process for each of the ten solutions, along with the formalized Lean proofs. The mathematical community will be scrutinizing these results in the coming weeks and months.

The bigger question is what happens next. If Astra can solve decade-old math problems for $2,000 in API costs, what else can it solve? And when it does ship publicly, how will developers, researchers, and businesses use a model designed to think for hours or days?

One thing is certain: the gap between "AI as a tool" and "AI as a researcher" just narrowed significantly.

Frequently Asked Questions

What is OpenAI Astra?

Astra is OpenAI's next major model family, designed for long-running, multi-agent reasoning tasks. An internal version has already solved ten previously unsolved mathematical problems. It has not yet been released to the public.

How much did it cost to solve the math problems with AI?

OpenAI says the tokens used to generate all ten solutions would have cost approximately $2,000 at current GPT-5.6 Sol API rates.

Were the AI-generated math proofs verified?

Yes. Each proof was formalized in Lean, a proof assistant that creates machine-checkable certificates of correctness. This allows independent verification without trusting OpenAI's claims.

Will Astra be released as GPT-6?

OpenAI has not decided whether Astra will ship as GPT-6 or as a variant within the GPT-5 line. There is no public release date yet.

Can I use AI models for mathematical research today?

Yes. Current frontier models like GPT-5.6 Sol, Claude Opus 5, and DeepSeek V4 Pro are available through the Qubax AI API and can assist with mathematical reasoning, proof verification, and research tasks.

What is the Leiden Declaration on AI and Mathematics?

The Leiden Declaration is a framework for how credit should be assigned in AI-assisted mathematical research. OpenAI referenced it when discussing the attribution of Astra's proofs.

Ready to build with frontier AI models? [Get started with Qubax AI](https://qubax.ai/docs) — unified API access to GPT-5.6, Claude, Gemini, DeepSeek, and 100+ more models with transparent pricing.