AI BriefingNVIDIA and Local AI Community Drive the Proliferation of Open-Source Models and Intelligent Agents
Key point
NVIDIA is partnering with the local AI community to expand the ecosystem of open models and agents.
Details
NVIDIA is highlighting partners, open-source communities, models, applications, and development tools that advance local AI throughout August. It supports developers and AI enthusiasts in building, customizing, and running AI agents locally using NVIDIA's open models and software, GPU-accelerated computing, libraries, and educational resources.
Key public models include:
- Cosmos 3 Edge: A 4-billion parameter model for robotics, autonomous driving, and vision AI, running on-device on NVIDIA DGX Spark and Jetson.
- MiniMax-H3: A 33-billion parameter model that generates stereo audio synchronized with video based on text, images, video, and audio. It supports ComfyUI and NVIDIA GPU-optimized checkpoints.
- Laguna S 2.1: A 118-billion parameter agentic coding model capable of performing tasks for hours. It can run on a single DGX Spark using NVFP4 checkpoints.
- DeepSeek-V4-Flash: A 284-billion parameter MoE model using 13 billion active parameters, offering a 1 million token context window. It can run on DGX Station via community-built GGUF versions.
- Inkling-Small: A multimodal model supporting text, images, and audio with adjustable reasoning intensity. It activates only 12 billion parameters per token out of 276 billion, running on one DGX Station or two DGX Sparks.
Unsloth Desktop was introduced as an open-source desktop application unifying local model inference and training, image and video diffusion, fine-tuning, agent integration, web research, and code execution. Alibaba's Wan-Animate-2 applies motion and facial expressions from motion videos to static character images, delivering high generation performance on NVIDIA RTX PRO 5000 Blackwell and RTX 5090.
Additionally, LTX-2.5 was released. It is an open video generation model targeting media and entertainment, robotics, and real-time generation, improving video quality, prompt adherence, and consistency in characters, scenes, and audio, while supporting multi-shot generation.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.