DownloadNVIDIA Nemotron 3.5 Bolt 74B-A9B BF16 is an open-weight, full-precision vision-language model designed for customization, post-training, and autonomous AI agent development. Compressed from Nemotron 3.5 Super VL using NVIDIA's Star Elastic method, it contains 74B total parameters with 9B active per token. Its hybrid Mamba-2–Transformer LatentMoE architecture integrates a C-RADIO v4-H vision encoder for processing text, images, and video. The model supports a 256K-token context window, structured tool calling, JSON output, and configurable reasoning modes. It is optimized for coding, browser automation, terminal operations, and multi-step agent workflows. BF16 weights preserve flexibility for supervised fine-tuning, reinforcement learning, distillation, and creating quantized variants. Deployment is supported through vLLM on NVIDIA GPUs, while LiLiCorr speculative decoding improves inference speed.
Features
- 74B total parameters with 9B activated per token
- Hybrid Mamba-2–Transformer LatentMoE architecture
- C-RADIO v4-H vision encoder for image and video understanding
- 256K-token context window for extended agent workflows
- Full BF16 precision for customization and post-training
- Supports supervised fine-tuning, reinforcement learning, and distillation
- Configurable reasoning, structured tool calling, and LiLiCorr decoding
- vLLM deployment on NVIDIA Ampere, Hopper, and Blackwell GPUs
Project Samples
