MiMo v2.6: Xiaomi Just Built the Best Open Model
Xiaomi just released MiMo v2.6, now the top open model on the Artificial Analysis index, and they livestreamed the entire RL run that produced it, which cost about $3.5 million. I test the Pro model inside the DeepSeek harness on agentic tasks with verification loops, including a self-researched landing page, a live ISS tracker, Blender assets, and Three.js scenes. I also break down the VRAM you'd actually need to run the Pro and Flash versions locally, and why "open" doesn't mean "runnable at home." Thanks to Firecrawl for making this video possible: firecrawl.link/promptengineer LINKS: Blogpost: https://mimo.xiaomi.com/mimo-v2-6 Live Run: https://mimo.xiaomi.com/rl/#overview Annoucement: https://x.com/XiaomiMiMo/status/2102138559952290106 Distilled Qwen-9B: https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B My voice to text App: whryte.com Website: https://engineerprompt.ai/ RAG Beyond Basics Course: https://prompt-s-site.thinkific.com/courses/rag Signup for Newsletter, localgpt: https://tally.so/r/3y9bb0 💻 Pre-configured localGPT VM: https://bit.ly/localGPT (use Code: PromptEngineering for 50% off). Signup for Newsletter, localgpt: https://tally.so/r/3y9bb0
