Xiaomi's MiMo Code Outperforms Claude Code on Long-Horizon Coding Tasks

Xiaomi's MiMo AI team has open-sourced MiMo Code V0.1.0, a terminal-native AI coding assistant that outperforms Anthropic's Claude Code on key agentic coding benchmarks, particularly on long-horizon, multi-step tasks (200+ steps). According to Xiaomi's internal beta release and survey of 576 developers, MiMo Code paired with MiMo-V2.5-Pro outperformed Claude Code paired with Claude Sonnet 4.6 on all three evaluations tested: SWE-bench Verified (82% vs. 79%), SWE-bench Pro (62% vs.

55%), and Terminal Bench 2 (73% vs. 69%). The MiMo Code project is a fork of the open-source OpenCode agent, extended with Xiaomi's own memory architecture, workflow modes, and model harness.

It features a cross-session memory system powered by SQLite FTS5 full-text search, which spans four layers: project memory, session checkpoints, scratch notes, and per-task progress logs. This system uses an independent 'checkpoint-writer' subagent to take notes, allowing the primary coding agent to focus on its work. Xiaomi also released MiMo-V2.5, its multimodal flagship model with a million-token context window, requiring no registration to get started.

The model is MIT-licensed and available on GitHub. MiMo Code is available now on GitHub under an MIT license and installs with a single terminal command on macOS and Linux or via npm on Windows. The company claims that its memory and state-management architecture pays off specifically on long-horizon work, citing a human double-blind A/B evaluation during its internal beta.

In this evaluation, covering 576 developers working in 474 real private repositories, MiMo Code's win rate rose above 65% for tasks with over 200 execution steps. MiMo Code offers easy integration with existing developer systems and voice control. It operates directly in the terminal, reading and writing files, running commands, and managing Git.

The tool requires zero configuration, connecting automatically to 'MiMo Auto,' a free-for-a-limited-time channel powered by Xiaomi's multimodal MiMo V2.5 model. The pricing for MiMo-V2.5 starts at $0.40 per million input tokens and $2.00 per million output tokens, while V2.5-Pro runs $1.00/$3.00 per million (input/output) up to 256K context. Xiaomi's aggressive pricing and open-source approach make MiMo Code an attractive option for developers.

Why this matters: The release of MiMo Code and its performance on long-horizon coding tasks has significant implications for the AI and development communities. It highlights the importance of memory and state-management architecture in agentic coding performance, an area where competitors are also focusing. For developers and businesses, MiMo Code offers a low-risk, potentially high-value evaluation candidate, with MIT-style licensing permitting modification and commercial integration.