NVIDIA open-sources post-trained multimodal Nemotron 3.5 Super VL on October 15, beating Claude Opus 5 and GPT-6.1 Sol on specialised benchmarks
NVIDIA on October 15 released and open-sourced the post-trained multimodal Nemotron 3.5 Super VL model, beating Claude Opus 5 and GPT-6.1 Sol on specialised benchmarks across medical imaging, radiology and industrial inspection; weights are published open-source with a focus on enterprise inference and on-prem deployment.
NVIDIA on October 15 released and open-sourced the post-trained multimodal Nemotron 3.5 Super VL model via Hugging Face and its enterprise platforms. The model beats Anthropic Claude Opus 5 and OpenAI GPT-6.1 Sol on specialised benchmarks across medical imaging, radiology and industrial inspection, with a focus on enterprise inference and on-prem deployment. It extends the Nemotron series' multimodal input capabilities with native image, video and text input; the post-training phase was carried out jointly by NVIDIA and partner hospitals and industrial enterprises. The model ships as open weights so customers can run inference inside their own data centres or clouds, avoiding sending sensitive medical or industrial data externally. Nemotron 3.5 Super VL fits an end-to-end stack together with NVIDIA NIM inference microservices, enterprise GPUs and the cuPotential CUDA-X library; NVIDIA positioned the release as core AI infrastructure for regulated industries such as healthcare, telecoms and manufacturing, putting it in direct frontier competition with Anthropic Mythos and the OpenAI GPT-6 family.