Kari Briski

NVIDIA released Nemotron 3.5 Super VL, a 120-billion-parameter open model designed for enterprise agentic workloads, alongside two edge-optimized variants, Bolt 74B-A9B 24GB and Bolt E4B 4GB. Super VL claims 5x faster throughput than comparable models while delivering accuracy comparable to NVIDIA's largest reasoning models at one-fifth the parameter count. The three models span deployment tiers from data centers to local devices, allowing organizations to run agents on their own infrastructure without trading capability for deployability.

NVIDIA released Nemotron 3.5 Lightning, a 30-billion-parameter mixture-of-experts model designed for specialized tasks in multi-agent AI systems, alongside NeMo Switchyard, an open source routing library. The model delivers up to 4x faster output speed and 30% faster agentic task completion compared to competitors in its class. Both tools enable enterprises to deploy customized AI across local systems, edge devices, and cloud infrastructure without rewriting applications.

NVIDIA released Nemotron 3 Nano Omni, an open multimodal model that unifies vision, audio, and language processing in a single system. The 30B-A3B hybrid mixture-of-experts architecture eliminates the need for separate perception models, delivering up to 9x higher throughput than comparable open omni models while maintaining responsiveness. The model tops six leaderboards for document intelligence and audio/video understanding, and is available immediately via Hugging Face, OpenRouter, and 25+ partner platforms.