Brief
Nemotron specialists reach IMO gold threshold with open natural-language pipeline
Two Nemotron 3 Ultra specialist checkpoints, trained with supervised fine-tuning and reinforcement learning, scored 30 out of 42 points at IMO 2026, the gold-medal threshold. The system works entirely in natural language, with no formal prover, external tools or internet access.
Starting from Nemotron 3 Ultra, the authors train two specialist checkpoints using supervised fine-tuning and reinforcement learning. An iterative search generates, verifies and refines candidate proofs, and a separate high-compute stage selects each final submission.
The system scored 30 out of 42 points at IMO 2026, reaching the gold-medal threshold, and operates entirely in natural language, with no formal prover, external tools or internet access.
The release covers 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.
Our reading
Our view is that the open release is the practical part: checkpoints, training data, code and a 200-problem benchmark let others run and extend the same natural-language proof pipeline without a formal prover.
What to do or watch
If you work on natural-language proof generation, use the released checkpoints, training data, code, submitted solutions, and Nemotron-IMO-Bench to reproduce and extend the pipeline. The unresolved question is whether independent users can match the 30-out-of-42 IMO 2026 result with these open artifacts.
Source details and supporting facts
Each line is stated by the page named above it.
Stated by arXiv
- The system scored 30 out of 42 points at IMO 2026, reaching the gold-medal threshold.
- The system operates entirely in natural language, with no formal prover, external tools, or internet access.
- Starting from Nemotron 3 Ultra, the authors train two specialist checkpoints using supervised fine-tuning and reinforcement learning.
- The release includes the two post-trained checkpoints, the training data, the training and inference code, the submitted solutions, and Nemotron-IMO-Bench.
- Nemotron-IMO-Bench is a new benchmark of 200 novel olympiad-level problems.
Sources
- arXivText stored 13 September 2026
How this story was checked. Written from the 1 page listed above, stored 13 September 2026; claims checked against that stored text on 14 September 2026.
What that means
- 5 of 5 reported statements were confirmed against the page that carries them; the rest were removed rather than published.
- Figures in the text were required to appear in the stored source text: yes. Identifiers: yes.
- The check reads stored text only: no claim rests on a fresh look that did not happen.
- Where the reporting was silent, the text says so instead of filling the gap.