Researchers have published an open recipe for reaching gold-medal performance in olympiad mathematics using natural-language proof generation. The work, released as a paper on Hugging Face, describes a test-time-compute pipeline built on Nemotron 3 Ultra that requires no formal prover, external tools, or internet access.
The team began from the Nemotron 3 Ultra base and trained two specialist checkpoints using supervised fine-tuning and reinforcement learning. They then studied how checkpoint choice, verification, and refinement affect proof quality for hard olympiad problems.
The resulting system combines three Nemotron 3 Ultra checkpoints — the general-availability model and the two post-trained specialists — in an iterative search that generates, verifies, and refines candidate proofs. A separate high-compute stage selects each final submission.
According to the paper, the system scored 30 out of 42 points at IMO 2026, reaching the gold-medal threshold. The authors also released the two post-trained checkpoints, the training data, the training and inference code, the submitted solutions, and Nemotron-IMO-Bench, a new benchmark of 200 novel olympiad-level problems.