NVIDIA's new Nemotron3 Ultra is defeated by Kimi K2.6 & GLM5.1 on coding tasks like TerminalBench, etc. In order to …
SemiAnalysis Twitter · SemiAnalysis (@SemiAnalysis_) · 2026-06-07
SemiAnalysis reports that NVIDIA's Nemotron3 Ultra is outperformed by Kimi K2.6 and GLM5.1 on coding benchmarks including TerminalBench, and calls on Jensen Huang to expand the Global Nemotron Coalition training committee to include frontier AI labs.
Appears in
Extraction
Topics: llm-benchmarksopen-source-modelsai-codingnvidia
Claims
- NVIDIA's Nemotron3 Ultra underperforms Kimi K2.6 and GLM5.1 on coding tasks including TerminalBench.
- NVIDIA should invite at least one frontier AI lab to the Global Nemotron Coalition training committee to produce stronger open models.
Key quotes
NVIDIA's new Nemotron3 Ultra is defeated by Kimi K2.6 & GLM5.1 on coding tasks like TerminalBench, etc.