Xiaomi Releases Open-Source MiMo Code AI Assistant to Compete with Anthropic’s Claude Code

Xiaomi has launched MiMo Code V0.1.0, an open-source, terminal-native AI coding assistant designed to handle long-horizon, multi-step software engineering tasks. Announced on June 10, 2026, the tool aims to address the common issue of AI “context amnesia” by utilizing a persistent, cross-session memory architecture. The project is currently available on GitHub under an MIT license, offering developers an alternative to existing agents like Anthropic’s Claude Code.

How MiMo Code Manages Long-Term Memory

Unlike standard coding agents that rely on increasingly compressed context windows, MiMo Code employs an explicit storage-and-retrieval system. According to Xiaomi’s technical documentation, the system uses SQLite FTS5 to manage four distinct memory layers: persistent project files (MEMORY.md), session checkpoints, scratch notes, and task progress logs. To prevent the primary AI from losing focus, a dedicated “checkpoint-writer” subagent runs in the background. This subagent acts like an architect, updating the project’s state and blueprints in real time so the primary agent can rebuild its environment accurately if the context window reaches its limit.

Performance Benchmarks and Agentic Capabilities

Xiaomi claims its agent harness offers a measurable advantage in complex software engineering tasks. In internal evaluations, the company reported that MiMo Code, when paired with its MiMo-V2.5-Pro model, achieved a 62% success rate on the SWE-bench Pro benchmark, compared to 57% when using the same model within the Claude Code harness. On Terminal Bench 2, the system recorded 73%, outperforming the 68% result from the Claude Code framework. While these figures suggest the agent’s architecture contributes a roughly five-point performance gain, these results are vendor-reported and have not yet been validated by independent third-party leaderboards.

Integration and Developer Workflow

MiMo Code is built to function as a drop-in replacement for developers already using terminal-based agents. The tool supports “Compose mode,” which allows users to set high-level goals for autonomous execution, including planning, coding, and testing. Additionally, the system features voice control integration via Xiaomi’s MiMo-ASR speech recognition technology. For developers looking to migrate, the CLI automatically imports existing configurations, including Model Context Protocol (MCP) servers and custom API keys from Claude Code environments.

Cost and Model Flexibility

A significant part of Xiaomi’s strategy involves bundling MiMo Code with “MiMo Auto,” a free, limited-time access tier for its flagship MiMo-V2.5 multimodal model. This model features a 1-million-token context window and follows a sparse mixture-of-experts architecture. For organizations that prefer other providers, the agent supports a “bring-your-own-model” approach. It is compatible with OpenAI-compliant APIs, allowing teams to connect to other frontier models from providers such as DeepSeek, Moonshot AI, and Zhipu AI.

Comparison of Market-Leading Coding Agent Costs

Considerations for Enterprise Adoption

Engineering leaders evaluating MiMo Code should weigh the benefits of its open-source, inspectable architecture against potential operational risks. While the MIT license allows for commercial integration, the “free” model access routes code context through Xiaomi’s servers, which may conflict with strict corporate data-residency policies. Furthermore, as a V0.1.0 release, the software is in its early stages of maturity. Organizations subject to U.S. government procurement restrictions regarding Chinese technology vendors should also review internal compliance requirements before deploying the tool in production environments.