OFICIAL NVIDIA Newsroom

NVIDIA and Local AI Community Fuel Open Source Models and Intelligent Agents

What happened
Based on NVIDIA Newsroom · Aug 11, 2026

NVIDIA and open source communities released new local AI models and tools in August, enabling developers to run advanced agents on consumer and edge hardware. The updates include robotics, video generation, coding and multimodal models optimized for NVIDIA GPUs.

🚩 Report a correction ↗ Open source

Video

Video available

Key points
·
The open source ecosystem is making it easier for AI enthusiasts and developers to build, customize and run increasingly capable agents locally.
·
Throughout August, NVIDIA is celebrating the partners and open source communities moving local AI forward, along with the models, applications and tools emerging across the ecosystem.
·
That includes NVIDIA’s latest open models, software and developer tools, plus the accelerated computing, libraries and educational resources that help users get started.
·
Follow along for the latest developments in this special-edition NVIDIA Local AI blog series, with new updates added over the coming weeks.
Key numbers
·
New models released include Cosmos 3 Edge, a 4-billion-parameter robotics and vision model; MiniMax-H3, a 33-billion-parameter video generation model; Poolside AI’s Laguna S 2.
·
Thinking Machines Lab introduced Inkling-Small, a 276-billion-parameter multimodal model with native reasoning across text, images and audio.
·
Alibaba released Wan-Animate-2, a 14-billion-parameter motion transfer model, and LTX-2.

NVIDIA highlighted August as a month for local AI progress, showcasing open models, software and tools designed to run on consumer and edge devices. The company emphasized accelerated computing resources, libraries and educational materials to help developers build and customize AI agents locally. Updates were shared through a special-edition blog series, with new entries planned over several weeks.

New models released include Cosmos 3 Edge, a 4-billion-parameter robotics and vision model; MiniMax-H3, a 33-billion-parameter video generation model; Poolside AI’s Laguna S 2.1, an 118-billion-parameter coding agent; and DeepSeek-V4-Flash, a 284-billion-parameter mixture-of-experts model with a 1 million-token context window. Each model is optimized for NVIDIA GPUs, including DGX Spark, Jetson and DGX Station systems.

Thinking Machines Lab introduced Inkling-Small, a 276-billion-parameter multimodal model with native reasoning across text, images and audio. Unsloth launched Unsloth Desktop, an open source desktop application for local model training and inference, integrating diffusion, fine-tuning and agent workflows. Alibaba released Wan-Animate-2, a 14-billion-parameter motion transfer model, and LTX-2.5, a state-of-the-art video generation model with multishot support and improved prompt adherence.

Meta unveiled Muse Glimmer, a 30-billion-parameter dense model optimized for coding and local agentic AI. The model supports over 200 tokens per second on NVIDIA RTX 5090 and is designed for always-on local agents, enabling private, multistep workflows without cloud dependency. Developers can fine-tune it using NVIDIA NeMo Automodel or run it with frameworks like vLLM and llama.cpp.

Original source → Deals on Clipraptor.com →