The development
What happened
NVIDIA's October report summarizes competition research documented in September. Separate Nemotron specialists and inference workflows reached scores above IOI and IMO gold thresholds; the IOI run was unofficial.
Key facts
NVIDIA reports 535.4/600 for Nemotron-3-Ultra-CC at IOI 2026, above the 361.12 gold threshold and outside the official ranking.
The separate IMO system scored 30/42 against a gold threshold of 29; official graders assessed its proofs.
Different specialists combined supervised fine-tuning, some reinforcement learning, and iterative generation, evaluation and refinement.
The authors provide checkpoints, data and recipes. The October post discusses papers first submitted in September.
The context
Why it matters
The work offers documented specialization and inference recipes for difficult coding and proof tasks. The scores belong to research systems with specific evaluation conditions and substantial training or inference work. They do not establish an official medal, low-cost reproduction or broad superiority of an unmodified model.