Nemotron 3.5 Super VL launches on SageMaker and joins the science suite
Understand this story
AWS announced that NVIDIA Nemotron 3.5 Super VL is now available on Amazon SageMaker JumpStart, deployable via SageMaker Studio or the Python SDK. The vision-language model uses a 120B-parameter MoE backbone, 12B activated parameters, and a C-RADIO vision encoder, supports up to 1 million tokens of context, and is offered in an NVFP4 quantized version and a BF16 full-precision version. New accounts have a default ml.g7 quota of zero and must first request a quota increase via Service Quotas; the JumpStart service configuration accepts only one image and one video per request, endpoints provide only base weights without loading LoRA adapters, and although checkpoints include an MTP head, speculative decoding is not enabled. NVIDIA subsequently launched Nemotron for Science, integrating the open-source Nemotron 3.5 Super VL model with scientific evaluations, the open chemistry dataset LitmusBench v0.2, and post-training recipes for researchers building domain science agents. NVIDIA says the model and dataset are now available on Hugging Face and optimized for local operation on 128GB NVIDIA DGX Spark systems.
AI overview based on source reports · Updated 1 days ago
Developments
2 developments
- Nov 17 00:00 · 1 reportsNVIDIA releases Nemotron for ScienceNVIDIA Newsroom:NVIDIA Introduces Nemotron for Science to Accelerate Discoveries With Agents
- Nov 10 04:01 · 1 reportsNVIDIA Nemotron 3.5 Super VL launches on SageMaker JumpStartAWS Machine Learning Blog:NVIDIA Nemotron 3.5 Super VL is now available in Amazon SageMaker JumpStart
Coverage timeline
Follow the coverage to explore different perspectives.
- NVIDIA BlogIan Buck Kicks Off NVIDIA’s Week at SC26 as Science Enters the Age of Agents
At SC26 in Chicago, NVIDIA's Ian Buck announced Nemotron for Science, an open toolkit combining the Nemotron 3.5 Super VL model with an open chemistry dataset and post-training recipes that researchers can run locally, alongside the Rubin HPC GPU and an expanded Vera CPU platform. He also detailed a quantum computing collaboration with IBM and Fermilab, a CUDA-Q decoding algorithm called TCBP-JOBS that cuts error-correction research computation from 12 weeks to 45 minutes, and a Photonics co-packaged optics switch deployment at Lambda. Rubin HPC, with more than 50,000 CUDA cores and support for FP64 to FP4, is expected in 2028.
- NVIDIA NewsroomNVIDIA Introduces Nemotron for Science to Accelerate Discoveries With Agents
NVIDIA released Nemotron for Science, combining its Nemotron 3.5 Super VL open model, post-training recipes and LitmusBench v0.2, an open chemistry dataset of nearly 50,000 questions for training and evaluating chemistry reasoning. Partners including Terray Therapeutics, Lila Sciences and Therna reported post-training gains, and institutions such as Argonne, KTH and RIKEN are using the models in scientific agent systems. The model and dataset are available now on Hugging Face, with the model also optimized to run locally on a 128GB DGX Spark system.
- AWS Machine Learning BlogNVIDIA Nemotron 3.5 Super VL is now available in Amazon SageMaker JumpStart
NVIDIA Nemotron 3.5 Super VL, a 120B-parameter vision-language model with 12B active parameters, a C-RADIO vision encoder, and up to 1M-token context, is now deployable in Amazon SageMaker JumpStart via SageMaker Studio or the SageMaker Python SDK. JumpStart offers an NVFP4 variant on G7 instances and BF16 reference weights on ml.p5en.48xlarge, with measured throughput of 63.07 tokens/sec on ml.g7.24xlarge and 140.85 tokens/sec for BF16 on ml.p5en.48xlarge. The model targets document intelligence, GUI understanding, video question answering, and multimodal agentic applications.
Activity for this story
- Current comparable activity
- 4
- Peak comparable activity
- 10Nov 17 01:00
- Change in the past 24 hours
- -49%
Trends compare the same fully observed participants, which may cover fewer participants than current activity. Move the pointer or tap the chart to inspect hourly activity; use the left and right arrow keys with a keyboard.