OpenAI's Astra model recently solved 10 major unsolved problems in mathematics and theoretical computer science—breakthroughs that would impress the academic community if a human mathematician had achieved them.
While the work is genuinely impressive, it has triggered an existential crisis among leading mathematicians questioning the future role of human mathematicians and the field's research and funding structures, especially given how rapidly AI capabilities in abstract math have advanced in just the past six to twelve months.
What happened
OpenAI's Astra model recently solved 10 longstanding problems in mathematics and theoretical computer science—including quantum game theory and sphere packing in higher dimensions—and published detailed documentation. The company initially claimed no progress had been made on these problems in the last 10 years, though one paper later revealed they had built heavily on work by two prior researchers, a credit OpenAI quietly changed after publication.
Why it matters
These are problems mathematicians genuinely care about and have spent years trying to solve; researchers told The Verge that if a human mathematician had solved any single one, it would likely secure an academic career. The speed of AI's shift from being unable to count letters in "strawberry" to solving professional-level abstract math in just six months to a year has triggered an existential crisis in the field about what mathematicians do and their future role, similar to what software engineering has grappled with for the past five years.
What to watch
The critical unknown is how many problems OpenAI attempted before achieving these 10 solutions—a figure the company knows but has not disclosed. The replicability of this feat and whether the models can solve another 10 remain unanswered, making it unclear whether this represents a sustained capability or a narrow breakthrough.
Ask the AI about this article →
The rapid shift in AI's mathematical capabilities represents a fundamental recalibration of what large language models can do. As recently as 2024, the conventional wisdom was that AI models were particularly bad at math—unable even to count letters in simple words like "strawberry." Yet within six to twelve months, models reached what researchers describe as professional-level performance on problems mathematicians have genuinely cared about and spent years trying to solve. This mirrors a pattern seen earlier in software engineering, where AI moved from being a curiosity to a tool that can handle significant professional work in a compressed timeframe.
The nature of the problems Astra solved offers clues to how this acceleration happened. The breakthrough problems are almost entirely self-contained theoretical puzzles where the work can be verified by running the proof again. There is no need for world knowledge or real-world intelligence; the system can apply horsepower and compute to a well-bounded problem space. This explains why AI excels at abstract mathematics while still being unable to reliably count or tell time—skills that require basic arithmetic or grounding in the external world, not high-level reasoning connections. Meanwhile, the mathematician community has noted that some areas, such as topology, may still be weak points for the models, confirming that the breakthrough is uneven across the discipline.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
Slack has introduced Slack Code, a feature that lets AI coding agents work in dedicated project channels where…
HUMAIN, the AI firm of Saudi Arabia's Public Investment Fund, has teased a new laptop developed with chip desi…
Chinese models—Moonshot's Kimi K3, Alibaba's Qwen3.8-Max, and GLM-5.3—have closed the performance gap with Ope…
Researchers at security firm Adversa discovered that Grok can be tricked into stealing user data—including nam…
Apple researchers introduced LINK, a method that improves cross-lingual knowledge transfer by randomly replaci…
Turing Award winner Richard Sutton, a founder of reinforcement learning, argued that synthetic data cannot sol…
The AI news that matters, in one minute each morning.
Sign up free