Xiaomi Open Sources Terminal AI Programming Assistant MiMo Code: Model Inference on Par with UltraClaude Code

According to Dyno-Sight Beating monitoring, Xiaomi has officially released and open-sourced the MiMo terminal AI programming assistant MiMo Code V0.1.0, using the permissive MIT open-source license. MiMo Code is based on the open-source project OpenCode with secondary development, integrating the time-limited free MiMo-V2.5 multimodal model and compatible with mainstream large-scale models APIs such as DeepSeek, Kimi, and GLM, as well as third-party Token Plans.

Traditional programming agents heavily rely on model-induced note-taking, often forgetting key context due to the models not actively triggering. MiMo Code introduces a persistent memory system, outsourcing state recording to independent sub-agents. When the session window approaches its limit, the sub-agents automatically save the state and reconstruct a clean brief for seamless handover to the main agent. The built-in /dream command runs automatically every 7 days, with an independent agent reading historical sessions and memory files, performing merge, deduplication, and path validity checks to compress scattered information into the current state for updating the global memory.

To enhance model and agent framework adaptability, MiMo Code has designed an exclusive Harness system specifically for the MiMo series models. Users can switch to Compose mode by pressing the Tab key, providing basic requirements for the system to autonomously execute the complete development loop of design, planning, coding, testing, and review. In authoritative tests such as SWE-Bench Pro and Terminal Bench 2, MiMo Code using the same base model achieved scores of 62% and 73%, respectively, improving by 5 percentage points compared to Claude Code.

MiMo Code also features voice control, allowing users to modify inputs or perform actions such as sending commands through verbal instructions. For deployment and usage, macOS and Linux users can install with a single curl command, while Windows users are supported to deploy using npm. Upon launch, the terminal will present a fully localized TUI interface, with a persistent status dashboard on the right for real-time progress monitoring.