Nvidia Nemotron 3.5
Nvidia's open-weight models for agents. Super VL targets enterprise agents, and Bolt runs local agents on PCs.
Announcement: blogs.nvidia.com ↗
Is it actually free? Free
Yes. The weights are free to download and run, but you still pay for your own hardware or hosting. Super VL is on Hugging Face, OpenRouter, ModelScope and NVIDIA NIM. Bolt 74B-A9B is on Hugging Face and works with llama.cpp, Ollama, LM Studio, Unsloth and vLLM.
Pricing
The open weights are free to download. Hosted prices depend on the provider.
Best for
- Super VL is a 120B mixture-of-experts model with 12B active parameters. It suits multimodal enterprise agents that you post-train on your own data.
- Bolt 74B-A9B 24GB is quantized to 2 bits and fits in 24GB of memory. It runs local agents on RTX and DGX Spark machines, at about 84 tokens per second on RTX Spark.
- Bolt E4B 4GB is for fast local chat and voice agents. Nvidia says it arrives Oct 23.
Not great for
Anyone without a capable GPU. The speed figures come from Nvidia.