OpenAI has announced its next-generation model family, Astra, which enables multiple agents to work together on complex problems for hours or even days. CEO Sam Altman has already demonstrated Astra to policymakers in Washington, but the company has not yet decided whether to release it as GPT-6 or a variant of GPT-5.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
OpenAI has officially confirmed its 'next-generation model family' Astra for the first time by publishing a report on solving ten long-standing open math problems. The move showcases AI's potential in scientific reasoning, but the model still needs to pass U.S. government review, and the reliability of multi-agent systems remains unresolved.
On August 1, 2026, OpenAI released a math report, officially confirming the name Astra for the first time, calling it the 'next-generation model family.' The report claims that an internal version of Astra solved ten open problems in mathematics and theoretical computer science, problems that mathematicians had not made progress on for at least a decade, and many for much longer.
These results span fields such as high-dimensional geometry, coding theory, group theory, quantum complexity, lattice cryptography, and extremal combinatorics. One of the proofs establishes the existence of non-sofic groups, solving a major open problem in group theory.
Mathematician Thomas Bloom of the University of Manchester called the results 'big news' on X, saying they are more significant than the counterexample to the unit distance conjecture published in May. However, he also pushed back against the idea that AI is replacing mathematicians, noting that AI is based on over a century of mathematical theory, built by mathematicians, and trained on everything mathematicians have written.
OpenAI stated that generating the tokens for all ten solutions would cost approximately $2,000 at Sol's API rates. After the model generated the arguments, humans collaborated with the same model to turn them into research papers, and each proof was formalized using Lean, producing machine-checkable certificates of mathematical correctness.
OpenAI also released demonstrations of the reasoning process for each solution. The company said that researchers helped prepare the papers and formalize the proofs, and that the company is responsible for their accuracy, but the mathematical arguments themselves come from Astra.
OpenAI believes that attributing fully AI-generated proofs to humans would misrepresent the system's contribution and genuine human intellectual work, and cited the 'Leiden Declaration on AI and Mathematics' as a reference for attribution in AI-assisted research.
According to The Information, the Astra model has been in testing and is expected to be the first model to undergo review under the new AI framework planned by the Trump administration, which requires AI models to be submitted to the federal government before public release. The government aims to finalize the framework by the end of this week.
OpenAI has not decided whether Astra will be released as a variant of GPT-6 or GPT-5 (such as GPT 5.7), and no release date has been set. CEO Sam Altman has already demonstrated Astra to policymakers in Washington.
A key question is whether the model can avoid error accumulation in long-running workflows and self-correct as context grows. This remains a major weakness of current agent systems. Multi-agent setups like Astra may perform worse on tightly coupled tasks (such as planning) because coordination overhead and error accumulation can offset benefits.
OpenAI's long-term goal is to build AI systems that can work continuously for hours or even days. Chief Scientist Jakub Pachocki said in a podcast last summer that the company wants models to plan, reason, and experiment over longer time horizons. OpenAI also plans to have fully autonomous AI researchers by March 2028, and AI systems at the level of research interns as early as September this year.
This article's information primarily comes from THE DECODER's report, with some details (such as Astra's testing status and regulatory plans) sourced from The Information, which are secondary sources and have not been independently verified. OpenAI's math report is an official release, but its claims about solving mathematical problems have not yet undergone peer review.
Astra's release marks a significant advance for OpenAI in scientific reasoning and long-duration task handling, but uncertainties around regulatory review, technical reliability, and release format mean its actual impact remains to be seen.
Primary report
Primary source