Xiaomi Open-Sources MiMo Code to Challenge Claude Code
The new MIT-licensed AI coding assistant hits 62% on SWE-bench Pro, powered by the MiMo-V2.5 model.
The landscape of self-hostable AI coding assistants just got a lot more interesting. In the early hours of June 11, Xiaomi officially released and open-sourced MiMo Code, a developer-focused tool designed to tackle real-world software engineering tasks.
Released under the highly permissive MIT license, MiMo Code arrives with a free, built-in MiMo-V2.5 model. For developers who prefer to keep their codebases local and avoid the vendor lock-in of proprietary APIs, this release offers a compelling new alternative in the self-hosted AI developer stack.
Shaking Up the Benchmarks
When a new coding assistant drops, the first question on every developer's mind is simple: how does it actually perform on real code?
According to initial release data, MiMo Code scored 62% on SWE-bench Pro. To put that into perspective, that is five percentage points higher than Anthropic's Claude Code. For those unfamiliar, SWE-bench is a rigorous benchmark that tests LLMs on their ability to resolve actual, multi-file GitHub issues in complex codebases. Scoring over 60% on the "Pro" variant suggests that the underlying MiMo-V2.5 model possesses a highly sophisticated understanding of context, dependency resolution, and codebase navigation.
While benchmarks should always be taken with a grain of salt until verified by independent local runs, beating a heavyweight like Claude Code on its own turf is a strong opening statement.
The Permissive Advantage
Many of the most capable AI coding tools on the market today are gated behind proprietary APIs, subscription models, or restrictive licensing terms. By opting for the MIT license, Xiaomi is positioning MiMo Code as an open infrastructure play.
Developers are free to modify, distribute, and integrate the tool into proprietary commercial workflows without worrying about licensing compliance. This makes it an attractive target for enterprise teams looking to build custom, internal AI coding pipelines where data privacy is paramount.
The Real Play: MiMo SoloEngine
While the developer community is understandably focused on the immediate utility of MiMo Code and its impressive SWE-bench Pro performance, industry observers point to a larger strategic move.
Behind the scenes, the real engine driving Xiaomi's AI developer ecosystem is MiMo SoloEngine. While MiMo Code serves as the highly visible, user-facing assistant that grabs the headlines, SoloEngine represents the foundational runtime and execution layer. It is this underlying architecture that will likely dictate how efficiently these models run on local developer hardware and how easily they can scale across diverse environments.
For developers looking to experiment with the latest in open-source AI engineering, the repository is now live. Whether MiMo Code can maintain its benchmark lead as other models iterate remains to be seen, but for now, the self-hosted AI stack has a powerful new contender.
Sources & further reading
Lenn writes about cloud platforms, Kubernetes internals, and the infrastructure decisions that quietly make or break engineering organizations. Based in Berlin's vibrant tech scene, they have a talent for turning dense platform-engineering topics into prose that people actually finish reading.
Discussion 0
No comments yet
Be the first to weigh in.