NVIDIA's Nemotron hits gold at both IOI and IMO: 535.4/600 at IOI, 30/42 at IMO
NVIDIAAI · x · 2026-10-14
NVIDIA and Hugging Face detail the post-training recipe behind Nemotron's gold-medal-level results at both IOI and IMO 2026, and release the competitive-programming specialist Nemotron-Labs-3-Competitive-Coding (based on Nemotron-3-Ultra) on Hugging Face.
- IOI 2026: Nemotron-3-Ultra-CC with SFT and feedback-driven GenCorrect inference scored 535.4/600, above the 361.12 gold threshold and the top human score of 498.27, in a live run under the same time/access/submission constraints as humans (unofficial, unranked)
- IMO 2026: general Nemotron 3 Ultra SFT/RL checkpoints in a generate-verify-refine system scored 30/42, above the official gold threshold of 29, graded by official IMO graders
- The recipe shows one foundation family adapting to two very different skill sets — time-constrained algorithmic coding and rigorous natural-language proofs — via SFT, RL, and feedback-driven inference
- The shared chart traces IOI progression: base 130 → SFT 262–280 → RL 291 → 468 after five TTC rounds, crossing bronze, silver, then gold thresholds
More from Models
- Hands-on with Gemini 4 Argon: real-time wildfire and tectonic viz built in 5 minutes — DavidmComfort · 2026-10-14
- Reddit rumor: Gemini 4 Boron to replace 3.1 Pro (unverified) — Any-Field-4793 · 2026-10-14
- Argon pushes Gemma past 120k tokens/sec on a single A100 with vLLM, a 2.9x speedup — osanseviero · 2026-10-14
- Anthropic merge data suggests new models go internal only ~5 weeks before release — ziv_ravid · 2026-10-14
- Perplexity reportedly prepping PPLX-Decider v1.2 27B decision model update this week — ChrisUniverse · 2026-10-14
- Anthropic bans Claude in China, Russia and 10 more countries under new regions policy — scaling01 · 2026-10-14