Models · Updated 7 Oct, 06:15 pm IST
Hugging Face’s Nemotron fine-tuned to gold-level at IOI and IMO 2026
Why it matters for readers: Provides concrete examples of what advanced fine-tuning and feedback loops enable in modern AI models.
- A Nemotron-3 variant (Nemotron-3-Ultra-CC) with supervised fine-tuning and GenCorrect scored 535.4/600 at an IOI 2026 run, above the gold threshold and the top human score reported.1
- A Nemotron 3 Ultra system using supervised fine-tuning, reinforcement learning, and a generate-verify-refine pipeline scored 30/42 at IMO 2026, above the official gold threshold of 29, with proofs graded by official IMO graders.1
- The IOI result was obtained in a live, prospective run under the same time, internet-access, and submission constraints as human contestants but was an unofficial, unsupervised benchmark and not included in official IOI rankings.1
- Hugging Face frames these results as evidence that Nemotron is a reusable foundation model that can be specialized to demanding domains via SFT, RL, and feedback-driven inference.1
Get a brief like this every morning
Uzha reads hundreds of sources and gives you the stories that matter for your work, with every source linked. Free.
Get started