Xiaomi MiMo Code Release

MiMo Code is a Terminal-Native AI Coding Assistant

Xiaomi has released MiMo Code, an open-source AI agentic coding harness designed to operate directly within the terminal. MiMo Code enables developers to read and write code, execute commands, manage Git repositories, and utilize a persistent memory system to maintain project context across multiple sessions.

Core Architecture and Features

MiMo Code is developed as a fork of OpenCode, inheriting its core capabilities including support for multiple providers, a Terminal User Interface (TUI), Language Server Protocol (LSP), Model Context Protocol (MCP), and a plugin system.

Beyond the base OpenCode functionality, Xiaomi has added several advanced agentic features:

  • Persistent Memory: A system for deep project understanding that evolves across sessions.
  • Autonomous Orchestration: Includes subagent orchestration and goal-driven autonomous loops.
  • Workflow Composition: Tools to compose complex workflows.
  • Self-Improvement: Implementation of "dream/distill" mechanisms for continuous improvement.
  • Context Management: Claims of "unlimited context" via automatic knowledge accumulation and lossless compression for large-scale projects.

Model Performance and Accessibility

Users have reported that the underlying MiMo models, particularly the MiMo-2.5 Pro and MiMo-2.5 Pro Ultraspeed, are highly competitive in terms of speed and capability. Some users have compared the performance to Claude Opus or Sonnet 4.6, noting that the models are particularly effective as "workhorses" for common coding tasks.

Accessibility is a key highlight of the release, as the tool can be installed and used without the friction of account creation or requiring a Chinese (+86) phone number, which is often a barrier for international users of Chinese software.

Technical Implementation and Community Feedback

Installation and Security Concerns

Community members have raised concerns regarding the official installation method for macOS and Linux, which suggests using curl -fsSL https://mimo.xiaomi.com/install | bash. Technical critics argue this is a dangerous practice for production environments, noting that the Windows installation via npm install -g @mimo-ai/cli is a more secure and standard approach to package management.

Open Source Integrity

Because MiMo Code is a fork of OpenCode, some developers have questioned why Xiaomi chose to create a separate clone rather than contributing improvements back to the upstream OpenCode project.

Token Usage and Billing

There are reports of significant discrepancies in how token usage is calculated and displayed. One user noted that while the harness showed 63.1K tokens for a task, the web-based usage progress bar indicated 152 million tokens, while a separate table on the same page showed 3 million tokens.

Summary of User Perspectives

The reception of MiMo Code is polarized between praise for Xiaomi's rapid advancement in frontier-level models and skepticism regarding the "open source" nature of the project's contributions:

"What a transformation by Xiaomi to build almost frontier level models. Five years back... they dint really bother about AI models and were using Baidu for NLP and vision."

"Today on open source vampires: xaomi forks an open source project, doesn't contribute upstream... and wants good PR."

"I think it would be hard to know if I was using Opus of MiMo if blindfolded in many instances."

Sources