ModelsPublished All news from that day
Ongoing case: Million-token context windows 6 stories
Nvidia released Nemotron 3.5 Super VL, a 120B open model taking images and video, with a 1M-token context.
Nvidia's developer blog of 15 October 2026 introduces Nemotron 3.5 Super VL, a 120-billion-parameter mixture-of-experts model that activates 12 billion per token and has weights on Hugging Face.
Check our sources · 7 facts from 2 sourcesNvidia published Nemotron 3.5 Super VL on 15 October 2026. Its developer blog describes a 120B-parameter mixture-of-experts architecture that activates 12B parameters per token. The model card lists text, images and video as inputs and text as output, with a context length of up to 1M tokens.
The model card says use is governed by the OpenMDW License Agreement, version 1.1, and that the model is ready for commercial or non-commercial use. The blog says the release includes base, post-trained and NVFP4-quantized checkpoints with training recipes, data and reinforcement-learning environments.
On its own benchmarks Nvidia reports 63.44% on Hermes Bench against 56.33% for Nemotron 3 Ultra. These are Nvidia's figures.
Underlined sentences link to the facts we checked. Dotted words open a short explainer.
Fact 2 of 7 · from NVIDIA Technical Blog, 15 Oct 2026
base, post-trained, and NVFP4-quantized checkpoints with training recipes, data, and reinforcement-learning environments
Fact 4 of 7 · from Hugging Face / NVIDIA
Input types: Text, images, and video; Output type: Text
Fact 6 of 7 · from Hugging Face / NVIDIA
OpenMDW License Agreement, version 1.1 ... This model is ready for commercial or non-commercial use.
Your reaction
We count reactions per story and day, never who reacted. The counts help us choose what goes in the monthly issue. If you are signed in, your own page shows yours too.
We count reactions per story and day, never who reacted. They help choose what goes in the monthly issue.
Check our sources
Every sentence above is checked against these 2 sources.
1 NVIDIA Nemotron 3.5 Super VL Powers Specialized Agents with Multimodal Reasoning
Open the source archived copy-
Nvidia's developer blog, dated 15 October 2026, describes Nemotron 3.5 Super VL as a 120B-parameter mixture-of-experts architecture that activates 12B parameters per token.
citeIts 120B-parameter mixture-of-experts (MoE) architecture activates 12B parameters per token
-
Nvidia's blog says the release includes base, post-trained and NVFP4-quantized checkpoints with training recipes, data and reinforcement-learning environments.
citebase, post-trained, and NVFP4-quantized checkpoints with training recipes, data, and reinforcement-learning environments
-
Nvidia's blog reports Hermes Bench scores of 63.44% for Nemotron 3.5 Super VL against 56.33% for Nemotron 3 Ultra; these are Nvidia's own figures.
citeOn Hermes Bench and LangChain behavioral evaluations, Nemotron 3.5 Super VL scores higher than Nemotron 3 Ultra. (table: 63.44% and 56.33%)
2 NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16 model card
Open the source archived copy-
The model card lists text, images and video as inputs and text as output.
citeInput types: Text, images, and video; Output type: Text
-
The model card gives a context length of up to 1M tokens.
citeUp to 1M tokens; deployment settings are listed below
-
The model card states that use of the model is governed by the OpenMDW License Agreement, version 1.1, and that it is ready for commercial or non-commercial use.
citeOpenMDW License Agreement, version 1.1 ... This model is ready for commercial or non-commercial use.
-
The model card gives a Hugging Face release date of 10/15/2026.
citeHugging Face - 10/15/2026
Every sentence above is checked against these 2 sources. Each fact has its own address you can cite.
Source 1 · NVIDIA Technical Blog · 15 Oct 2026
NVIDIA Nemotron 3.5 Super VL Powers Specialized Agents with Multimodal Reasoning
Nvidia's developer blog, dated 15 October 2026, describes Nemotron 3.5 Super VL as a 120B-parameter mixture-of-experts architecture that activates 12B parameters per token.
Its 120B-parameter mixture-of-experts (MoE) architecture activates 12B parameters per token
Nvidia's blog says the release includes base, post-trained and NVFP4-quantized checkpoints with training recipes, data and reinforcement-learning environments.
base, post-trained, and NVFP4-quantized checkpoints with training recipes, data, and reinforcement-learning environments
Nvidia's blog reports Hermes Bench scores of 63.44% for Nemotron 3.5 Super VL against 56.33% for Nemotron 3 Ultra; these are Nvidia's own figures.
On Hermes Bench and LangChain behavioral evaluations, Nemotron 3.5 Super VL scores higher than Nemotron 3 Ultra. (table: 63.44% and 56.33%)
Source 2 · Hugging Face / NVIDIA
NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16 model card
The model card lists text, images and video as inputs and text as output.
Input types: Text, images, and video; Output type: Text
The model card gives a context length of up to 1M tokens.
Up to 1M tokens; deployment settings are listed below
The model card states that use of the model is governed by the OpenMDW License Agreement, version 1.1, and that it is ready for commercial or non-commercial use.
OpenMDW License Agreement, version 1.1 ... This model is ready for commercial or non-commercial use.
Show all 4 factsShow fewer facts
Topics
The morning email
On the mornings we publish: the three top stories and up to four short ones. Free.
We email you a link to confirm. An issue may include one sponsor, always labelled Sponsored · Advertisement. Our emails count opens and clicks, not who made them. Unsubscribe in one click. What we keep