AI BriefingNVIDIA Releases Nemotron-Labs-3-Competitive-Coding-550B, First AI to Outscore Top Human on IOI Problem Set
Key point
The 550B-parameter model scored 535.4 out of 600 on the IOI 2026 problem set, surpassing the top human contestant's score of 498.27.
Details
NVIDIA has released Nemotron-Labs-3-Competitive-Coding-550B-A55B-NVFP4, a specialized model fine-tuned from NVIDIA-Nemotron-3-Ultra-550B-A55B for competitive programming tasks. The model was trained for one epoch on 477,642 synthetic reasoning traces distilled from GLM-5.2 across 22,000 curated problems from 16 regional and international contest families.
Performance on IOI 2026
The model was evaluated live and prospectively on the IOI 2026 problem set under official contest constraints, including time limits and internet access rules. Using GenCorrect, an iterative closed-loop test-time compute strategy that refines solutions based on evaluator feedback, the system achieved a score of 535.4 out of 600.
This result exceeds both the gold-medal threshold of 361.12 and the top human contestant's score of 498.27, marking it as the first AI system reported to outscore the highest-scoring human contestant on an IOI problem set.
Technical Details
- Base Model: NVIDIA-Nemotron-3-Ultra-550B-A55B
- Distillation Source: GLM-5.2 (selected for higher accuracy and ~30% shorter generations compared to a DeepSeek-V4-Flash variant)
- Training Data: 477,642 synthetic reasoning traces
- Problem Scope: 22,000 problems across 16 contest families
- License: Ready for commercial and non-commercial use
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Stanford and NVIDIA Release 'Contrastive Language Model' CLM with 9x Faster InferenceInception Releases Ultra-Fast LLM 'Mercury 2.5' Generating 770 Tokens Per Second