Nvidia's 550B Nemotron beats top human coder at IOI 2026

Nvidia researchers report that their 550-billion-parameter Nemotron-3-Ultra-CC scored 535.4 out of 600 at IOI 2026, above the top human contestant's 498.27 and well above the gold threshold of 361.12.

The claim comes from a new arXiv preprint by Aleksander Ficek and colleagues, describing an end-to-end pipeline of curated problems, synthetic reasoning traces, supervised fine-tuning, and reinforcement learning. "To our knowledge, this is the first AI system to outscore the highest-scoring human contestant on an IOI problem set," the authors write.

Two models were trained on 22,000 curated problems: Nemotron-3-Nano-CC (30B-A3B) with both SFT and RL, and Nemotron-3-Ultra-CC (550B-A55B) with SFT alone. A test-time strategy called GenCorrect "iteratively generates, evaluates, and refines diverse solutions." On IOI 2025, Nano-CC went from 130 points to 291 after post-training, then to 468 with GenCorrect, exceeding the gold threshold of 438.3. Ultra-CC reached 502.

The IOI 2026 run was prospective. The paper says the competition-specific Ultra-CC system operated "under the same time, internet-access, and submission constraints as human contestants." It names no human contestant, publishes no per-problem breakdown, and gives no compute budget.

This is a research claim from Nvidia's own team about its own model, not an independently adjudicated result.