Xiaomi’s latest AI coding tool claims to outperform Claude Code on complex tasks

Xiaomi has launched Mimo Code v0.1.0, an open-source AI coding assistant designed for developers working in the terminal. The tool reportedly surpasses Anthropic’s Claude Code on benchmarks like SWE-bench Verified (82% vs. 79%), SWE-bench Pro (62% vs. 55%), and Terminal Bench 2 (73% vs. 69%), according to Xiaomi’s internal tests. The company attributes these gains to Mimo Code’s agentic architecture, including persistent memory and workflow design, rather than the underlying AI model alone. Xiaomi’s MiMo V2.5 multimodal AI powers the assistant, which is currently free and supports voice commands, Git operations, and integration with providers like OpenAI, Kimi, and DeepSeek. Mimo Code is built on Xiaomi’s OpenCode agent framework, enhanced with memory systems and workflow modes. It can execute commands, manage Git repositories, and maintain project knowledge across sessions. The tool also learns from past interactions, converting repetitive tasks into reusable skills and improving efficiency over time. It features experimental parallel reasoning and multi-agent collaboration for complex coding projects. The assistant is available on GitHub under an MIT license, installable via terminal commands on MacOS and Linux, or npm on Windows. Xiaomi’s claims are based on a double-blind evaluation involving 576 developers across 474 private repositories, showing stronger performance on long-horizon tasks compared to Claude Code. The company highlights its ‘infinite context’ capability, which stores and compresses key project information to aid decision-making. Mimo Code’s architecture includes structured workflows for planning, building, and validating code, along with self-review features. It supports multiple AI models and integrates with Claude Code tools, ensuring compatibility with existing workflows. The assistant’s voice recognition and persistent memory further enhance its utility for developers managing large-scale projects.