NVIDIA's Fine-tuned Nemotron Beats Top Human Contestants at IOI

NVIDIA's fine-tuned Nemotron scored 535.4/600 on the IOI 2026 problem set, surpassing top human contestants. The model was trained via a pipeline combining curated problems, synthetic reasoning data, SFT, RL, and iterative test-time optimization.

2026-09-03 ~ 2026-09-05 · 2 related posts