NVIDIA's Nemotron coding model scores 535.4/600 on IOI 2026, beating top human, now on Hugging Face
NVIDIAAI · x · 2026-10-06
NVIDIA announced that its fine-tuned Nemotron competitive-programming model exceeded the gold medal threshold on the IOI 2026 problem set, scoring 535.4 out of 600 as graded by the IOI team — higher than the top-scoring human contestant. The unofficial run took place in Uzbekistan under supervision of the International Technical Committee. The model, Nemotron-Labs-3-Competitive-Coding (550B-A55B, NVFP4), is a specialist built on Nemotron-3-Ultra and is now available on Hugging Face.
More from Models
- Bittensor guard model gains 8 F1 points in 4 weeks to near-SOTA via miner attacks — bittingthembits · 2026-10-06
- Nat Lambert: Reflection Joins Nvidia, Thinking Machines in Falling Behind Chinese LLMs — natolambert · 2026-10-06
- User burns $107 in 90 minutes on Claude Code Cloud, billed like an API key — MicahBerkley · 2026-10-06
- Claude remains ~2.5x more efficient in token/plan comparison, blogger claims — kimmonismus · 2026-10-06
- Beam's benchmarks disappoint: US open model lags Chinese rivals like DeepSeek, says Reddy — bindureddy · 2026-10-06
- Model Census pings every listed LLM daily to track availability and deprecations — repligate · 2026-10-06