OpenAI claims its Astra model solved 10 difficult math problems; Anthropic says Fable solved 5
OpenAI is making a strong headway. In a recent blog post, the AI startup claims that its Astra model—not yet released to the public—has achieved 10 major breakthroughs in mathematics. However, an Anthropic employee asserts that Claude Fable has also managed to solve 5 of these problems on its own.
OpenAI has published a blog post claiming to have achieved 10 significant breakthroughs in mathematics. These achievements were the work of Astra, OpenAI's new AI model that has not yet been officially released. However, Anthropic has responded by noting that Claude Fable is already capable of solving 5 of those same problems.
In the post titled "Ten breakthroughs in mathematics and theoretical computer science," OpenAI indicates that it has solved or made substantial progress on 10 long-standing open mathematical problems. All of this, the startup adds, was "achieved by an internal version of Astra, our next major model."
The results listed by OpenAI cover areas ranging from coding theory and operator algebras to quantum complexity, lattice-based cryptography, and extremal combinatorics.
Shortly after OpenAI's post, Anthropic engineer Levent Alpoge responded. In a message on X, Levent claimed to have solved 5 of those same problems using Claude Fable—a model already available to the public—in less than 24 hours.
"So, after 24 hours, I have half of them solved with Fable. I didn't see much discussion about the prompts in the announcement, but the setup is similar to the one I used, for example, in my announcement regarding the unit distance problem," he commented.
Levent added that Fable was able to solve these five problems completely autonomously after receiving generic instructions. Furthermore, the AI did not have access to the Internet. The problems it reportedly solved with Fable include arithmetic circuit complexity, quantum parallel repetition, and the closest vector problem.
Levent Alpoge’s post comes at a time when competition between OpenAI and Anthropic continues to intensify. Both companies are investing billions of dollars in an effort to gain an edge in the AI race. Additionally, both AI startups are expected to go public before the end of this year, potentially reaching valuations of one trillion dollars.
What can OpenAI’s Astra do?
Returning to Astra: although OpenAI explains that the total number of tokens used to find these solutions would cost around $2,000 (approximately 190,000 rupees) based on API rates, the startup offers no further details about its new AI model.
In other words, we do not know if Astra is part of the GPT-5.6 family or if it might mark the beginning of GPT-6. However, Astra appears to be more powerful than GPT-5.6. This could make Astra the most advanced model in existence if it is launched soon.
In its blog post, OpenAI stated that the emergence of systems capable of contributing to mathematical research raises questions that a technology company cannot resolve on its own.
The AI startup declared that it respected concerns regarding the impact of AI on mathematics. The company added that attribution must reflect how the result was obtained, noting that attributing human authorship to a proof generated entirely by an AI system would misrepresent both the system's contribution and human intellectual work.
According to reports, OpenAI CEO Sam Altman was in Washington, D.C., last week showcasing Astra to federal officials.
It is worth noting that OpenAI previously revealed that two of its models—GPT-5.6 Sol and an even more capable pre-release model—attempted to hack the database of the U.S. company HuggingFace during an evaluation test. It is possible that Astra was the pre-release model mentioned in that instance, although there has been no confirmation from OpenAI.
