Xiaomi says its new AI coding model beats Claude Code
What's the story
Xiaomi has unveiled MiMo Code v0.1.0, an open-source terminal-native AI coding assistant for developers. The model has outperformed Anthropic's Claude Code in key benchmarks related to agentic coding, especially long-horizon and multi-step tasks. Long-horizon tasks are goals given to an AI agent that may take many sequential steps, decisions, and hundreds of actions to complete a task.
Features
Installation and features of MiMo Code
MiMo Code v0.1.0 can read and write code, run commands, manage Git repositories, and keep project memory across terminal sessions. Xiaomi has released the model on GitHub under an MIT license. It can be installed with a single terminal command on macOS and Linux or via npm on Windows. The tool is designed to handle large software projects with its 'infinite context' through automatic knowledge storage and compression.
Adaptability
Model can turn repetitive tasks into reusable skills
MiMo Code comes with features that let it learn from past interactions. It can save useful project knowledge, turn repetitive tasks into reusable skills, and improve over time. The model also has speech recognition support for voice command interaction. It works with multiple AI model providers such as Anthropic, OpenAI, Kimi, DeepSeek, and GLM and is compatible with Claude Code tools and workflows.
Efficiency
It also supports multiple AI agents working on tasks
MiMo Code features a persistent memory system, supports multiple AI agents working on tasks, tracks complex task hierarchies, and offers experimental features for parallel reasoning and agent switching. According to Xiaomi's internal benchmarks, MiMo Code outperformed Claude Code in several software engineering tests when paired with their respective flagship AI models. The company conducted a double-blind study with 576 developers across 474 private repositories.
Superiority
Similar performance on shorter tasks
Xiaomi claims that MiMo Code and Claude Code performed similarly on shorter tasks. However, for projects with over 200 execution steps, MiMo Code's win rate was above 65%. This advantage was possible due to long-term memory and state-management capabilities designed for complex, multi-session coding tasks. The company did not publish comparisons with OpenAI's Codex or Google's Gemini CLI.