MiMo Code Goes Open Source: Xiaomi’s AI Coding Model Released
Originally published on Xiaomi / Hacker News — June 12, 2026
Summary
Xiaomi released MiMo Code as an open-source AI coding model, joining the growing ecosystem of developer-focused LLMs. MiMo Code scored 71 in aggregate benchmarks and 65.8 in agentic coding evaluations, outperforming DeepSeek V4 Flash (57 aggregate, 49.1 agentic) according to the MiMo V2.5 model comparison data, though DeepSeek V4 Flash remains competitive on pure coding tasks at a lower price point.
Key Details
- MiMo Code is now open-source and publicly available, bringing Xiaomi into the AI coding tools space.
- The MiMo V2.5 model scores 71 aggregate / 65.8 agentic vs DeepSeek V4 Flash at 57 / 49.1.
- DeepSeek V4 Flash is cheaper at $0.28/M output tokens vs MiMo V2.5 at $2/M.
- MiMo is positioned as a reasoning model optimized for agentic tasks.
- The open-source release enables community contributions and self-hosting.
Read Original
Curated by Brain Bot for Abhay’s KB — June 12, 2026