Daily News · 1 min read

Hugging Face AI Updates: August 5, 2026

hugging-face open-source edge-ai agents video-generation multimodal

1. Liquid AI Releases LFM2.5-2.6B for On-Device Agents

Hugging Face. Liquid AI published LFM2.5-2.6B on the Hugging Face Hub, a 2.6 billion parameter model built on the LFM2 architecture with a 128K context window, pre-trained on roughly 34 trillion tokens and post-trained with agentic reinforcement learning. The company reports it is competitive with models four times its size on tool use, instruction following, and multi-step agentic tasks, running at 220 tokens per second on an Apple M5 Max while using under 2.5GB of memory. Source

2. MiniMax Open-Sources H3 Omni-Modal Video Model on the Hugging Face Hub

Hugging Face. MiniMax published open weights for H3, the omni-modal generation model behind its Hailuo line, to the Hugging Face Hub. The 33 billion parameter release jointly processes text, images, video, and audio to generate 4 to 15 second videos up to 2K resolution with native synchronized stereo audio, and ships under the MiniMax H3 Community License, which restricts commercial use for larger organizations and public use in several jurisdictions. Source