NVIDIA broadens local AI push with open models and agent tools

NVIDIA is using an August Local AI series to highlight open models, tools and community projects that make it easier to build and run capable agents on local hardware. Recent releases include Cosmos 3 Edge, a 4-billion-parameter open world model for robotics, autonomous vehicles and vision AI; MiniMax-H3, a 33-billion-parameter open weights model for video and synchronized stereo audio; Poolside AI’s Laguna S 2.1, a 118-billion-parameter agentic coding model; and DeepSeek-V4-Flash, a 284-billion-parameter MoE model with 13 billion active parameters and a 1 million-token context window.

Creative and multimodal tools are also expanding. Thinking Machines Lab’s Inkling-Small is a 276-billion-parameter multimodal model that activates just 12 billion parameters per token, while Alibaba’s Wan-Animate-2 is a 14-billion-parameter model for transferring motion and facial expressions from driving video to static characters. LTX-2.5 adds multishot video generation, generative edits, stronger prompt adherence and NVIDIA RTX optimizations that deliver up to 20% faster performance and 40% memory savings on an NVIDIA RTX 6000 PRO GPU.

Meta’s Muse Glimmer and NVIDIA’s Nemotron 3.5 Lightning target always-on local agent workflows. Muse Glimmer is a 30-billion-parameter open weight model with a 120K+ context window that runs at over 200 tokens per second on RTX 5090. Nemotron 3.5 Lightning is a customizable open 30B MoE model that delivers up to 4x faster token generation and 30% faster time to completion compared with open models in its class. NVIDIA Sync updates add clustering, remote access and monitoring for DGX Spark systems, while NeMo Switchyard routes agent workflow steps across models based on accuracy, speed and cost.