NVIDIA models

We track 21 NVIDIA models, 20 of them with open weights. The most recent release is Switchyard, released 21 Sept 2026.

Official site · Hugging Face

ModelReleasedContextInput $/1MOutput $/1MWeights
Switchyard21 Sept 20261M–––
Nemotron 3.5 Lightning 30B A3B11 Aug 2026262K$0.05$0.17Open
Nemotron Ultra4 Jun 2026203K$0.50$2.20Open
Nemotron 3.5 Content Safety (free)4 Jun 2026128K$0.20$0.20Open
Nemotron 3 Nano Omni (free)28 Apr 2026256K––Open
nemotron-3-content-safetydeprecated16 Apr 2026128K––Open
llama-nemotron-rerank-vl-1b-v231 Mar 2026128K––Open
nemotron-voicechat16 Mar 2026128K––Open
Nemotron Super11 Mar 2026203K$0.08$0.45Open
llama-nemotron-embed-vl-1b-v210 Feb 202633K––Open
nemotron-content-safety-reasoning-4bdeprecated22 Jan 2026128K––Open
Nemotron 3 Nano 30B A3B15 Dec 2025262K$0.05$0.20Open
Nvidia Nemotron Nano 12B V2 VL28 Oct 2025131K$0.20$0.60Open
llama-3.1-nemotron-safety-guard-8b-v328 Oct 2025128K––Open
Nvidia Nemotron Nano 9B V218 Aug 2025131K$0.06$0.23Open
Nvidia Nemotron Super 49B8 Aug 2025128K––Open
Llama 3.3 Nemotron Super 49B v1.5deprecated25 Jul 2025131K––Open
mistral-nemotrondeprecated11 Jun 2025128K––Open
Llama 3.1 Nemotron 70B Instruct15 Apr 2025128K––Open
Llama 3.1 Nemotron Ultra 253B7 Apr 2025128K––Open
nemotron-mini-4b-instructdeprecated21 Aug 2024128K––Open