NVIDIA has developed a post-training pipeline for its Nemotron-3 language models, enabling them to achieve gold-medal level performance in the International Olympiad in Informatics (IOI). The approach combines large-scale problem curation, synthetic reasoning traces, supervised fine-tuning, and reinforcement learning.

  • Nemotron-3-Nano-CC (30B-A3B) was trained with supervised fine-tuning and reinforcement learning, while Nemotron-3-Ultra-CC (550B-A55B) used supervised fine-tuning alone.
  • The GenCorrect strategy iteratively generates, evaluates, and refines diverse solutions during test time.
  • On IOI 2025, Nano-CC scored 468 points with GenCorrect, exceeding the gold threshold of 438.3, while Ultra-CC reached 502.
  • A competition-specific Ultra-CC system evaluated prospectively on IOI 2026 constraints scored 535.4 out of 600, surpassing the top human score of 498.27.

This marks the first time an AI system has outscored the highest-scoring human contestant on an IOI problem set under standard competition conditions.