IT Home, October 15: NVIDIA has announced the launch of Nemotron 3.5 Super VL, an open multimodal model designed for AI agents requiring continuous operation and execution of multi-step tasks.
The model employs a Mixture of Experts (MoE) architecture with a total of 120 billion parameters and 12 billion activated parameters. NVIDIA states that, when executing agent tasks on a single GPU, this model achieves up to 5 times the throughput of other open models of similar scale.
Nemotron 3.5 Super VL integrates reasoning, instruction following, and tool-calling capabilities with text, image, document, and video understanding in a single model, enabling applications such as code testing and debugging, cross-site information retrieval, document analysis, cybersecurity threat detection, and video content search. Developers can use the NVIDIA NeMo toolkit to fine-tune the model with their own data, tools, policies, and workflows, and customize its behavior to meet specific business needs.
NVIDIA also disclosed several enterprise customization cases. For example, AI data technology provider Scale AI developed a supplier credit processing agent for media and entertainment company Skydance Corporation, with expected reductions of 97.5% in inference costs and 80% in latency for the associated workloads.
The financial operations platform Ramp is converting receipt images into structured JSON data containing merchant, date, and payment information, expected to increase throughput by up to 50% and reduce latency by 90%. AI21 is collaborating with the AI security company Alice to customize models for threat intelligence and vulnerability analysis, improving accuracy on related tasks by 40%.
In model evaluation, NVIDIA reports that Nemotron 3.5 Super VL scored 80.8% on the CharXiv chart reasoning benchmark, surpassing Nemotron 3 Nano Omni’s 63%; when combined with Python code interpretation, the score reached 84%.
On the Hermes Bench and LangChain behavioral evaluations, the model achieved scores of 63.44% and 87.05%, respectively, surpassing Nemotron 3 Ultra’s scores of 56.33% and 84.3%. NVIDIA also stated that the model is among the top-performing open models on the OCRBench v2 leaderboard.
To reduce inference overhead, Nemotron 3.5 Super VL employs techniques such as speculative decoding, mixed precision, and context reuse. NVIDIA states that its LiL token issuance speculative decoding method can increase token generation speed by up to 2.3x on DGX Spark. The model also supports NVFP4 quantization, using 4-bit weights and 4-bit activations on NVIDIA Blackwell architecture; on non-Blackwell platforms, the same checkpoint uses 4-bit weights and 16-bit activations.
NVIDIA also unveiled deployment optimization solutions for distributed inference. According to its disclosed data, under test conditions using 16 NVIDIA B200 GPUs and generating 50 tokens per user per second, NVIDIA Dynamo, combined with cache-aware routing, session affinity, and KV cache offloading, can serve up to 8 times more users while reducing first-token latency by 85%.
This release also includes the base model, post-trained model, and NVFP4 quantized model weights, along with training recipes, datasets, and reinforcement learning environments.
NVIDIA stated that Nemotron 3.5 Super VL used approximately 26 trillion tokens during pre-training and continued pre-training, around 3.8 million samples during post-training, and underwent reinforcement learning across dozens of environments. The newly released dataset covers areas such as agent tool usage, financial reasoning, instruction following, long-context understanding, fairness, and vision-language understanding.
In addition to Super VL, NVIDIA has launched two Nemotron 3.5 Bolt edge models. The Nemotron 3.5 Bolt 74B-A9B 24GB features a MoE architecture with 74 billion total parameters and 9 billion activated parameters, compressed to 24GB via native 2-bit quantization, designed for local agent tasks and compatible with llama.cpp, vLLM, Ollama, Unsloth, and LM Studio. The Nemotron 3.5 Bolt E4B 4GB is targeted at local voice assistants, interactive characters, and embedded interfaces, with a planned release on October 23.
IT Home notes that Nemotron 3.5 Super VL and Bolt 74B-A9B 24GB are now available for access; developers can obtain the model resources through channels such as Hugging Face, OpenRouter, and ModelScope. Super VL can be deployed on DGX Spark, DGX Station, RTX systems, data centers, or in the cloud; Bolt E4B 4GB is scheduled for release on October 23.