NVIDIA Nemotron 3.5 Super VL
Nemotron 3.5 Super VL is a 120B-parameter (12B active) open-weight vision-language model from NVIDIA, released on October 15, 2026. It takes text, image and video input and returns text; NVIDIA says it is ready for commercial use.
Released
October 15, 2026
Type
multimodal
License
open-weight
OpenMDW License Agreement, version 1.1
Context
—
NVIDIA states up to 1M tokens of context.
Capabilities
Architecture
Parameters: 120B total, 12B active (Mamba-2 and Transformer hybrid mixture of experts)
Sources
Every figure on this page was checked against these pages on . A figure we could not source is left out.
- NVIDIA build: nemotron-3.5-super-vl-120b-a12b model card (opens in a new tab)
- Model card: nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4 (opens in a new tab)
- NVIDIA blog: Nemotron 3.5 Super VL (opens in a new tab)
Spotted something out of date? Tell us.
Links
In the news: NVIDIA Nemotron 3.5 Super VL and NVIDIA
MarkTechPost
NVIDIA Releases Nemotron 3.5 Super VL, a 120B Open Multimodal Agent Model
NVIDIA has released Nemotron 3.5 Super VL, an open multimodal agent model designed for always-on enterprise agents. This 120-billion-parameter mixture-of-experts model activates 12 billion parameters per token and can process text, images, documents, and video. The model is capable of reasoning, calling tools, and writing code within a single architecture. For developers, this open checkpoint replaces the need for separate text and vision models. NVIDIA also introduced two edge models: Nemotron 3.5 Bolt 74B-A9B and Bolt E4B. Nemotron 3.5 Super VL features a context window of up to 1 million tokens and supports various hardware configurations, including DGX Spark, H200, H100, and GB200 with an NVFP4 checkpoint. A BF16 checkpoint is also available. NVIDIA claims the model offers up to 5x higher single-GPU throughput compared to open peers, with strong performance in reasoning, agentic coding, and document vision. Its performance on the GPQA-Diamond benchmark is 89.90 with the NVFP4 checkpoint. The model's weights are distributed under the OpenMDW-1.1 license. NVIDIA has also provided open datasets, training recipes, and reinforcement learning environments in NeMo Gym. The model card specifies text, image, and video inputs, with text output, and video input is limited to MP4 clips up to two minutes. The architecture interleaves Mamba-2, mixture-of-experts, and attention layers, with each MoE layer containing 512 routed experts. A C-RADIO vision encoder and a connector process visual inputs into the language backbone. Speed enhancements include multi-token prediction, LiLiCorr for faster token generation on DGX Spark, NVFP4 with quantization-aware distillation for efficient processing, and context reuse for prefix caching.
NVIDIA Newsroom
Adobe and NVIDIA Expand Collaboration to Speed Agentic Creation
Adobe and NVIDIA are expanding their partnership to optimize Adobe's flagship creative tools for faster AI-powered creation, production, and personalization. The collaboration will enhance tools such as Adobe Photoshop, Premiere, Substance 3D, Frame.io, and Brand Concierge. This expanded collaboration aims to deliver accelerated AI-powered workflows for creative professionals.
NVIDIA Newsroom
Open From Data Center to Desktop: NVIDIA Expands Nemotron Family With Highest Efficiency Super VL and Bolt Models
NVIDIA has launched Nemotron 3.5 Super VL, an open 120-billion-parameter model designed for specialized always-on agents, offering fast task completion and high accuracy. The company also introduced two new edge models: Nemotron 3.5 Bolt 74B-A9B 24GB, which balances intelligence and speed for local agentic workloads, and Nemotron 3.5 Bolt E4B 4GB, optimized for similar tasks.
MarkTechPost
Liquid AI Releases LFM2.5-24B-A2B: A 2B-Active MoE for On-Device Agents
Liquid AI has released LFM2.5-24B-A2B, an open-weight reasoning model for agents designed to run on laptops and PCs. This 24B-parameter Mixture-of-Experts (MoE) model activates about 2B parameters per token and is the final member of the LFM2.5 family. The model is trained within popular agent harnesses, focusing on multi-step tool use rather than open chat. Base and post-trained weights are available on Hugging Face. LFM2.5-24B-A2B features a 131,072-token context window and can run on laptops and mini PCs using Apple M5 Max or AMD Ryzen AI Max+ 395 processors via GGUF/MLX. It also supports single NVIDIA B300 for serving and offers FP8, GGUF, ONNX, and MLX 4/5/6/8-bit builds. In performance tests, LFM2.5-24B-A2B leads or matches larger MoEs on five out of eight agentic and tool-use benchmarks. It achieved 27.69 on AutomationBench, more than triple the score of Qwen3.6-35B-A3B. However, it performed less strongly on SWE-Bench Verified (OpenCode) with a score of 51.00 compared to Qwen3.6-35B-A3B's 66.20. The model is text-only and reasoning-only, built for local agent loops, with each assistant turn beginning with a <think> block. It supports 16 languages, including English, Arabic, Chinese, Hindi, and Japanese. Its architecture retains LFM2’s 40-layer hybrid structure, incorporating 30 double-gated short convolution blocks and 10 grouped query attention (GQA) blocks. Each MoE block contains 64 experts with top-4 routing. Context increased from 32,768 to 131,072 tokens, and vocabulary doubled from 65K to 128K via an in-place tokenizer expansion for improved non-Latin tokenization. Training involved scaling pre-training from 17T to 33T tokens, followed by mid-training with 2T tokens at a 32K context and then 300B tokens at a 131K context, focusing on tool calls and multi-turn interactions.
NVIDIA Newsroom
NVIDIA Commits $250 Million to Advance UK Agentic Safety and Security Research Over the Next Five Years
NVIDIA announced a five-year compute donation valued at $250 million to strengthen the United Kingdom’s agentic safety and security research capability. This commitment aims to advance research in AI safety and security within the UK.