NVIDIA Nemotron 3.5 Bolt 74B A9B 24GB
nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB
Open Source · chat · open-weights
Open
Alert me on changes
Context
256K
Max output
—
Weights
Open
API $/1M
—
Modalities
text · image
Released
07 Oct 2026
License: other · nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB
AI summary
● machine-written
NVIDIA Nemotron 3.5 Bolt 74B-A9B released as open-weights model
NVIDIA Nemotron 3.5 Bolt 74B-A9B is a 74-billion-parameter mixture-of-experts model with 9 billion active parameters per token, compressed to 24GB using 2-bit quantization with quantization-aware training. The model supports text and image inputs via a C-RADIO v4-H vision encoder and is released as open-weights for download, fine-tuning, and offline use. It runs on standard hardware via llama.cpp, vLLM, Ollama, Unsloth, and LM Studio.
What's new
- Open-weights release with training data and recipes published
- Supports up to 256K tokens context length
- 74B total parameters compressed to 24GB using 2-bit quantization
- Multimodal: text and image input via C-RADIO v4-H vision encoder
- Compressed from Nemotron 3.5 Super VL 120B-A12B
Best for
Offline agent inference on consumer hardwareFine-tuning for custom applicationsPrivacy-sensitive workloads requiring on-device processingLong-context text and image analysis tasks
Source: https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB