Nemotron 3.5 Bolt
Compact open vision-language MoE for running agents locally.
Nemotron 3.5 Bolt is NVIDIA’s smaller open vision-language model, a hybrid Mamba-Transformer mixture of experts (74B total, 9B active) compressed from Nemotron 3.5 Super VL. NVIDIA publishes a full-precision checkpoint and a 24GB build sized for a single workstation GPU or DGX Spark. It reads text, images and video, and thinking can be switched on or off.
- Provider
- NVIDIA
- Type
- Language model
- Released
- Oct 15, 2026
- Context window
- 262K tokens
- Input
- text, image, video
- Output
- text
- Open weights
- Yes
- License
- OpenMDW 1.1
Best for
- Agents
- Vision
- Local & on-device
- Long context
Strengths
- Fits in 24GB (quantized build)
- Image and video input
- Open weights
More from NVIDIA
- Nemotron 3 Super — NVIDIA. Open hybrid Mamba-Transformer MoE for multi-agent applications.
- Nemotron 3.5 Lightning — NVIDIA. Small open 30B-A3B model tuned for low latency.
- Nemotron 3.5 Super VL — NVIDIA. Open multimodal MoE for documents, GUIs and video.
- Nemotron Labs Dory 1.0 4B — NVIDIA. Small open looped model that reuses a shared block to think harder.
- Nemotron 3 Ultra — NVIDIA. NVIDIA’s largest open model: hybrid Mamba-Transformer MoE, 55B active.
- Nemotron 3 Embed 1B — NVIDIA. Small open embedding model for high-throughput enterprise retrieval.
- Cosmos 3 — NVIDIA. NVIDIA’s open omni-modal world foundation model for physical AI.
- Isaac GR00T N1.7 — NVIDIA. Open 3B cross-embodiment VLA for humanoids and manipulators.