NVIDIA releases Nemotron 3.5 Super VL and Bolt models for agentic workloads across data center and edge
Nemotron 3.5 Super VL is an open 120B-parameter model optimized for high-volume agentic tasks, while Nemotron 3.5 Bolt models target local deployments with 24GB and 4GB variants.
1 source · cross-referenced
- NVIDIA released Nemotron 3.5 Super VL, an open 120-billion-parameter model optimized for agentic tasks with up to 5x faster throughput than peers in its class.
- Two new edge models, Nemotron 3.5 Bolt 74B-A9B 24GB and Bolt E4B 4GB, are designed for local agentic workloads on NVIDIA hardware.
- Super VL and Bolt 74B-A9B are available now; Bolt E4B launches on October 23, 2026.
- The models are open-weight and customizable, with NVIDIA providing training recipes, datasets, and NeMo integration for post-training.
NVIDIA announced the release of Nemotron 3.5 Super VL, an open 120-billion-parameter model designed for high-volume agentic tasks such as instruction following, tool use, and multistep business workflows. The model achieves frontier-level accuracy while being roughly one-fifth the size of NVIDIA’s largest reasoning model, according to the company. On the Artificial Analysis Intelligence Index leaderboard, Super VL delivers up to 5x faster throughput compared to other models in its class.
The company also introduced two new edge-focused models: Nemotron 3.5 Bolt 74B-A9B 24GB and Nemotron 3.5 Bolt E4B 4GB. Bolt 74B-A9B is optimized for local agentic workloads on NVIDIA RTX/DGX Spark systems, using 2-bit quantization to fit within 24GB of memory. It is designed to balance intelligence and speed for everyday agentic tasks, such as web research, writing, and code debugging, without relying on cloud processing. Bolt E4B 4GB targets fast conversational agents and physical agents, supporting use cases like local voice assistants and embedded interfaces on GeForce RTX, Jetson, and RTX/DGX Spark hardware.
Super VL and Bolt 74B-A9B are available immediately, while Bolt E4B is scheduled for release on October 23, 2026. The three models collectively cover agent deployments from large enterprise servers to edge devices and local hardware, enabling organizations to choose an optimized model for their infrastructure tier.
NVIDIA emphasizes the openness and customizability of the Nemotron 3.5 family. The models ship with open weights, open datasets, and training recipes, allowing organizations to post-train them on proprietary data, tools, and workflows. This approach supports domain-specific customization for tasks such as financial document analysis, cybersecurity, and medical claims processing. NVIDIA NeMo provides Python-based pipelines for post-training, and the company has released open datasets on Hugging Face spanning agentic tool use, finance, instruction following, long context, fairness, and vision-language understanding.
The models are integrated into NVIDIA’s software ecosystem, including NVIDIA NIM for production-ready microservices and Dynamo for serving up to 8x more users with 85% lower latency. Super VL is available on Hugging Face, Open Router, ModelScope, and as an NVIDIA NIM package, and runs across NVIDIA’s hardware stack, including DGX Spark, DGX Station, and RTX systems.
NVIDIA highlights adoption of Super VL by organizations such as Scale AI, EY, AfterQuery, AI21, Alice, Mercor, and ModelScope, which are post-training the model for specialized tasks like vendor credit finance, M&A due diligence, credit agreement analysis, cybersecurity resilience, and coding benchmarks. Additional integrations include Atlassian, CrowdStrike, Dassault Systèmes, Perplexity, Salesforce, and Synopsys, which are customizing the model for agentic workflows in their respective domains.
- Nov 11, 2026 · Hugging Face
Hugging Face proposes token pooling-aware training to improve compressed multi-vector retrieval
Trust79 - Nov 11, 2026 · TechCrunch — AI
Fractile raises $720M Series C to develop AI chips and compiler stack
Trust79 - Nov 11, 2026 · The Verge — AI
Google’s Tabletop AI experiment lets users build custom collaborative apps via waitlist
Trust79