NVIDIA AI Updates: October 16, 2026

1. NVIDIA Releases Nemotron 3.5 Super VL, a 120B Open Multimodal Model for Agents

NVIDIA. NVIDIA released Nemotron 3.5 Super VL, an open 120B-parameter mixture-of-experts model with 12B active parameters per token that combines reasoning, tool use, and text, image, document, and video understanding in one model. NVIDIA reports up to 5x higher single-GPU throughput on agentic tasks than other open models in its size class, and says it was trained on about 26 trillion tokens and post-trained with roughly 3.8 million samples plus reinforcement learning. It is available on Hugging Face, OpenRouter, ModelScope, and as an NVIDIA NIM, alongside open datasets and NeMo Gym RL environments, and partners including Scale AI, Ramp, AI21, Atlassian, CrowdStrike, Perplexity, and Salesforce are post-training or integrating it. The smaller Nemotron 3.5 Bolt 74B-A9B is now available, with Bolt E4B (a 4GB edge model) due October 23. Source

2. Meta’s Muse Spark Open Ships With NVFP4 Checkpoints and a NIM for Blackwell Ultra

NVIDIA. NVIDIA detailed deployment of Meta’s Muse Spark Open, an 816B mixture-of-experts model with 42B active parameters that was trained on GB300 NVL72 Blackwell Ultra systems. Meta-generated NVFP4 checkpoints use native FP4 Tensor Core acceleration to cut memory footprint and free room for larger KV caches in long-context sessions. NVIDIA reports up to 50K tokens per second per GPU and up to 600 tokens per second per user on GB300 NVL72, and offers a downloadable NIM for production deployment. Source

3. NVIDIA Joins European Commission’s AI in Science Enablers Initiative

NVIDIA. At the AI in Science Summit in Dublin, the European Commission launched the AI in Science Enablers Initiative, tied to the RAISE (Resource for AI Science in Europe) program, and NVIDIA joined with commitments covering open models and tools, compute access, training, and startup support. The NVIDIA Deep Learning Institute launched an open source courseware hub with openly licensed course materials that require no enrollment or fees. NVIDIA is also working with national AI factories in France, Spain, and Sweden to give Inception startups and SMEs access to sovereign European compute. Source

4. NVIDIA Outlines Nordic AI Deployments, Including the Sferical One Supercomputer

NVIDIA. NVIDIA described its activity in the Nordics, including investments in Swedish startups Lovable and Legora and a projected 1 gigawatt of AI cloud capacity in the region by the end of 2027. A Swedish consortium anchored by Ericsson, AstraZeneca, Saab, SEB, and Wallenberg Investments launched Sferical One, a shared supercomputer on two DGX SuperPODs with 1,152 GPUs that will build services on Nemotron open models. Other examples include Spotify using TensorRT-LLM for podcast recommendations, Elisa launching a beta of the Nemotron-based meeting assistant Moontalk AIRI, and Swedish speech lab Klang fine-tuning Parakeet v3 into a Swedish transcription model it says reduces errors by 75%. Source

5. NVIDIA and AI4Bharat Announce a 2.8 Trillion Token Indic Pretraining Corpus

NVIDIA. NVIDIA and IIT Madras’s AI4Bharat lab introduced Nemotron-pretraining-Indic-v1, a corpus of about 2.8 trillion tokens covering all 22 languages listed in India’s Constitution, with native-script data for all 22 and romanized text for twelve. The data is translated from curated English sources including math, code, and question-answer content, with quality assessment and repair steps that preserve equations, tables, and code. The corpus is “releasing soon” together with the open GPU-accelerated generation pipeline, including prompts and quality checks. Continued pretraining for Nemotron 3.5 Super VL used 10 billion Indic tokens drawn from a twelve-language subset plus natural Indic text. Source