A new study introduces the Station, an open-world environment where AI agents from different model families collaborate on mathematical research without a central coordinator. The system allows agents to choose their own directions, run experiments, and build a shared literature.

Across 12 problems from the Alpha Evolve catalogue and two additional case studies, the Station produced novel results on five problems. These include a new infinite family of finite-field Kakeya sets, exact 604-point kissing configurations in dimension 11, and improved bounds for the discretized Kakeya needle and sign uncertainty problems.

The research also achieved a substantially improved lower bound for Erdős's minimum-overlap problem. Beyond numerical findings, the agents generated theorems and explanatory analyses, making their results interpretable and useful for future work.

All agent dialogues, proofs, and verification code are released publicly, offering a transparent record of the discovery process. This marks a step toward autonomous scientific exploration in mathematics.