Xiaomi has recently introduced an advanced AI coding agent known as MiMo Code V0.1, which is designed to outperform existing competitors like Anthropic’s Claude Code. This new tool, launched on June 11 under an open-source MIT license, excels in handling complex coding tasks that require multiple steps, making it highly useful for developers engaging with real-world projects.
MiMo Code V0.1 emerges from the MiMo-V2.5 model family and proves itself in intricate, multi-step challenges. The tool has notably outperformed Claude Code in scenarios that involve over 200 steps and more than 1,000 tool calls, while also utilizing significantly fewer tokens. This efficiency translates to lower operational costs for developers who implement it.
#How Does MiMo Code Compare on Key Benchmarks?
In evaluating MiMo Code’s capabilities, the Terminal-Bench 2.0 benchmark illustrates its superiority, where it scored 86.7%, far exceeding Claude's 65.4%. This reflects a substantial improvement, highlighting a 21-percentage-point advantage that underscores the tool's efficiency and accuracy.
Furthermore, in the SWE-bench Pro tests, MiMo Code achieved a score of 57.2%, showcasing its ability to effectively resolve real-world issues listed on GitHub, further establishing its reliability and technical prowess.
#What Makes MiMo Code Efficient?
The MiMo-V2.5-Pro model is central to this performance gain, featuring a mixture-of-experts architecture that encompasses approximately 1.02 trillion parameters. A significant benefit of this architecture is its 1-million-token context capacity, which enables the coding agent to maintain coherent performance across lengthy tasks. Other tools tend to lose focus as task complexity increases, but MiMo Code’s design ensures continuity and relevance throughout multifaceted coding challenges.
Another crucial feature of MiMo Code is its persistent memory capability, which allows developers to retain project context across various sessions. This means that developers can seamlessly pick up their work without needing to revisit prior explanations, making their workflow more efficient.
The terminal-native design allows MiMo Code to operate directly within the developer's existing terminal environment. This integration reduces friction, keeping developers aligned with their natural work processes. Additionally, Xiaomi has conveniently included API support for Claude Code, enabling developers to leverage specific tasks using Claude without needing to switch between tools.
#What is Xiaomi’s Vision for AI in Coding?
Xiaomi’s journey in AI development has been steady, with earlier versions like the MiMo V2-Flash model paving the way for current advancements. By April 2026, daily token usage on the MiMo platform surpassed 1 trillion tokens, indicating a growing dependency and trust in this technology.
The open-source nature of MiMo Code under the MIT license empowers anyone to download, modify, and deploy it at no cost, further inviting innovation and collaboration within the coding community.