What's this about?
NVIDIA had its open-weights model Nemotron 3 Ultra sit the 2026 International Mathematical Olympiad problems under contest conditions — same time limit, no internet, no tools. The IMO team graded its solutions 30/42, above the 29-point gold-medal threshold.
What exactly is the "IMO" and why does gold matter?
The International Mathematical Olympiad is the world's most prestigious math competition for pre-university students; gold medals go to roughly the top 8% of the world's best young mathematicians. When closed models (unreleased Google/OpenAI systems) first hit gold level in 2025 it was a landmark. This is the first time a model whose weights anyone can download claims that level.
What's the community saying?
1.1k likes on NVIDIA's post; commentary led by Ethan Mollick
- Mollick's point: the open-closed capability gap on elite reasoning is now about one year — and he suspects Kimi K3 and GLM-5.2 would also qualify if tested.
- Caveats raised: self-administered and graded on NVIDIA's initiative — not an official IMO entry — so the "same conditions" claim rests on NVIDIA's setup.
- It feeds the week's US–China open-weights story: the strongest open models are suddenly a geopolitical topic.