NVIDIA Launches Open LLMs for Enterprise and Edge
NVIDIA has released Nemotron 3.5 Super VL, a 120-billion-parameter open model, alongside two new Bolt models targeting edge and local deployments. These models are designed for agentic and multimodal workflows, with open weights, datasets, and training recipes available to all users.
What’s New: Super VL and Bolt Models
- Nemotron 3.5 Super VL: 120B parameters, open, optimized for high-throughput agentic tasks, and claims 5x faster task completion compared to peers.
- Bolt 74B-A9B 24GB: Edge model for local agentic workloads, fits in 24GB memory with 2-bit quantization, runs on RTX/DGX Spark and similar hardware.
- Bolt E4B 4GB: Compact model for conversational and physical agents, deployable on Jetson and GeForce RTX devices, available Oct 23.
Evidence and Mechanism
- Super VL and Bolt 74B-A9B are available now; Bolt E4B launches Oct 23.
- Open weights, datasets, and recipes enable full customization and on-premises deployment.
- Organizations across industries are already post-training Super VL for specialized tasks.
Implications for Indie Builders
Independent developers can now deploy high-performance agentic models locally, avoiding cloud lock-in and retaining full control over data and customization. The open release also supports fine-tuning for niche applications.
Practical Judgment
While NVIDIA’s performance claims are strong, real-world results will depend on independent benchmarks. Early adopters should test these models in their own workflows and monitor community feedback for practical guidance.