OpenAI’s New Astra Model Sparks a Crisis Over the Future of Mathematics

OpenAI announced an internal version of its next‑generation model, Astra, and released a “scorecard” showing ten new research results in high‑dimensional geometry, coding theory, group theory, arithmetic‑circuit complexity, quantum complexity, lattice cryptography, and extremal combinatorics.

Among the highlighted achievements are:

Construction of a non‑sofic group.

Disproof of the Connes rigidity conjecture.

Resolution of three Erdős problems.

Advancement of a recent vector problem closely tied to post‑quantum cryptography.

OpenAI research scientist Noam Brown noted that the team did not exhaustively invest effort on every problem; the token cost for the model to find all solutions was roughly $2,000, averaging about $200 per problem.

The announcement triggered strong reactions. Mathematician Jay Cummings, author of *Proofs: A Long‑Form Mathematics Textbook*, called the results “incredible.” Toronto professor Daniel Litt admitted he had underestimated AI’s capabilities and suggested that publishing Annals‑level number‑theory work with large models is only a matter of time.

At the same time, many researchers expressed a deeper existential unease: will mathematicians still have work? Is the discipline still worth pursuing? Some argue that mathematics’ true value lies in the human experience of discovery, not merely in producing proofs. Kirwin Hampshire’s essay “Mathematics in the Dark Night” likens the situation to Borges’s Library of Babel, warning that if AI can instantly generate countless elegant proofs, the act of creation itself may lose its meaning.

Other voices see a different picture. One commentator argued that AI will not replace mathematicians but will free them from tedious formalism, allowing focus on intuition and strategic judgment. Brown also clarified that the AI has not yet solved open problems, created new branches, or proposed interesting conjectures, so substantial work remains for human mathematicians.

Overall, Astra’s ten results have irrevocably shifted the conversation from “Can AI do mathematics?” to “What are humans still pursuing in mathematics now that AI can handle many proofs?” The article ends by inviting readers to reflect on this emerging crisis.