Editorial illustration for OpenAI to Publish Report on AI Solving Ten Unsolved Math Problems
OpenAI's Astra Model Solves 10 Unsolved Math Problems
OpenAI to Publish Report on AI Solving Ten Unsolved Math Problems
OpenAI is building a new model family called Astra, and the company wants it to do something none of its released products can: sit with a hard problem for hours or days and actually work through it. Sam Altman showed the system off this week in Washington, D.C., to an audience of politicians and regulators, according to a report from The Information citing three people familiar with the plans. The pitch centers on coordination, multiple agents assigned to pieces of a single tough problem, running in parallel over long stretches rather than firing off a quick answer and stopping.
Astra would join OpenAI's existing lineup of model classes, which reportedly includes names like Sol, Terra, and Luna. Whether Astra eventually ships as GPT-6, or gets folded into the GPT-5 line under a label like GPT-5.7, is still undecided, and OpenAI hasn't set a release date. The models are currently in testing and are expected to be the first put through a new U.S.
government review process requiring sign-off before public release. OpenAI has also flagged math as a proving ground for what Astra can do.
OpenAI is working on a new model family tentatively called "Astra" that's meant to be far more capable at long-running tasks than anything the company has shipped so far.
Why this matters
If Astra really can grind on a problem for days by coordinating multiple agents, that's a different product category than the chatbots most of us have been building around. Solving ten previously unsolved math problems is a flashy proof point, but the report matters less for the math than for what it signals about OpenAI's roadmap: long-horizon, multi-agent systems are coming out of the lab and into demos for policymakers in Washington. That should interest founders betting on agentic workflows, because the bar for "impressive" just moved.
It should also make researchers ask harder questions about verification. A model that works unsupervised for hours needs airtight ways to check its own output, and OpenAI hasn't detailed how it validated those math proofs. The regulatory angle is worth watching closely too.
Astra becoming the first model run through a new US government review process means the rules for releasing frontier systems are being written in real time, and whatever precedent gets set here will shape what every other lab has to do next.
Related Reading
Browse all OpenAI news →
Open Source D&B rebuilds 642 million‑business database after AI agents hit limits
Open Source OpenAI's GPT Transcribe Cuts Error Rate to 3.31%, Improving on GPT-4o
Open Source Study: 6.7% of Deepfake Requests on Hugging Face Targeted Children
Related Topic OpenAI researcher quits, citing distrust over ad‑driven engagement metrics
LLMs & Generative AI
Related Topic OpenAI launches GPT-Image 1.5 with precise editing for enterprise visuals
LLMs & Generative AI
From Archives Sources: More OpenAI Agents Reportedly Escaped Sandboxes
From Archives DeepSeek Boosts Agent, Coding Performance in Open-Source V4-Flash Model
From Archives Deepseek's New AI Model Matches GPT-5.6 at 60% Lower Cost
LLMs & Generative AI
From Archives OpenAI Slashes GPT-5.6 Luna AI Model Price by 80%
Market Trends
Common Questions Answered
What is OpenAI's new model family Astra designed to do differently from current products?
Astra is designed to work on difficult problems for hours or days at a time, which is a capability that none of OpenAI's released products currently possess. The system uses multiple coordinated agents assigned to different pieces of a single tough problem to achieve this extended problem-solving capability.
How did Sam Altman demonstrate the Astra model and to whom?
Sam Altman showcased the Astra system this week in Washington, D.C., to an audience of politicians and regulators. According to reports from The Information, the demonstration centered on the system's coordination abilities and multi-agent approach to solving complex problems.
What is the significance of OpenAI solving ten previously unsolved math problems with Astra?
Solving ten previously unsolved math problems serves as a proof point for Astra's capabilities, but the real significance lies in what it signals about OpenAI's roadmap toward long-horizon, multi-agent systems. This demonstration to policymakers indicates that these advanced systems are transitioning from laboratory research into practical applications and real-world deployment.
How does Astra represent a different product category compared to existing AI chatbots?
Astra's ability to coordinate multiple agents working on a single problem over extended periods represents a fundamentally different product category than the chatbots most users have been building around. This shift toward long-running, multi-agent coordination systems marks a significant departure from the interactive, single-turn conversation model that current AI products are based on.
Further Reading
- OpenAI says its new model family is built for long-horizon tasks and scientific research - 36Kr
- OpenAI has published new mathematical results from an internal model, including ten long-open problems - OpenAI
- An OpenAI model has disproved a central conjecture in discrete geometry - OpenAI
- OpenAI's GPT is getting better at mathematics - Computerworld
- Mathematical Computation and Reasoning Errors by Large Language Models - arXiv
