SemiAnalysis: Kimi K3 Beats NVIDIA's Nemotron, Calls Committee Model a Failure

FinanceYF5 · x · 2026-07-26

SemiAnalysis reports that the Kimi K3 model significantly outperforms NVIDIA's Nemotron 3 Ultra.

The analysis criticizes Jensen Huang's "committee" approach for developing Nemotron, arguing that this single-coalition model fosters groupthink. It also highlights how a single member's failure (like Mistral's pre-training misstep) can drag down the entire committee, and notes that members are reluctant to share their best ideas, datasets, and benchmarks.

They argue that true open-source progress stems from competition among independent labs based on free-market principles. To maintain leadership in open-source AI, the US should adopt the current Chinese AI lab model: independent teams testing diverse approaches and iterating rapidly, rather than relying on a restricted committee structure.

Related event: SemiAnalysis: Kimi K3 Beats Nemotron(4 posts)→

Original post →

More from Companies & People

Companies & People channel →