Xiaomi has announced MiMo Code v0.1.0, a new open source AI programming assistant for software developers. This tool is designed to handle large-scale, long-term projects and is capable of overcoming Claude Code of Anthropic in some complex programming tasks.
According to Xiaomi, MiMo Code not only supports writing and editing source code but can also execute commands in terminals, manage Git storage and maintain project memory between different working sessions.
This platform is built on OpenCode, an open source framework, and Xiaomi also adds a long-term memory system, workflow management mechanism and integration capabilities with its own AI model.
MiMo Code operated by MiMo V2.5, Xiaomi's multi-functional AI model is currently provided free of charge for a limited time.
The model owns a context window of up to one million tokens, allowing processing large amounts of project data and information in one working session.
One of the highlights of MiMo Code is the ability to memorize project knowledge for a long time. According to Xiaomi, the system can automatically store, compress and access important information, helping AI maintain the context when working on projects that last weeks or months.
In addition, this tool is also designed according to a structured working process. AI can self-plan, build technical specifications, perform tasks, check results and create reports.
The system also has the ability to learn from previous interactions, turning repeated tasks into skills that can be reused in the future.
MiMo Code supports voice control and is compatible with many different AI models such as OpenAI, Anthropic, DeepSeek, Kimi or GLM. This helps developers flexibly choose AI platforms that suit their usage needs.
Regarding performance, Xiaomi said that MiMo Code achieved 82% on the SWE-bench Verified test, higher than Claude Code's 79%. On the SWE-bench Pro evaluation standard, this tool achieved 62% compared to 55% of its competitor. In the Terminal Bench 2 evaluation, MiMo Code achieved 73%, slightly higher than the 69% that Claude Code recorded.
However, these results were self-announced by Xiaomi, so they need to be viewed cautiously. The company also did not make comparisons with other major competitors such as Codex of OpenAI or Gemini CLI of Google.
Some public data shows that Codex CLI uses GPT-5.5 at about 82% on Terminal Bench 2, which is higher than what MiMo Code reports.
However, the emergence of MiMo Code shows a new trend in the field of programming AI. Instead of just focusing on upgrading language models, technology companies are increasingly focusing on memory systems, planning capabilities, process management and AI agent coordination. These are considered factors that can make a difference in large-scale real-world software development projects.
