Xiaomi Corp (1810.HK) announced on June 11 the launch of MiMo Code, an experimental AI programming assistant marking the company’s first entry into the Coding Agent sector. The product extends the MiMo model’s capabilities from core AI to agent-level applications, building a “model + agent” ecosystem.
Earlier this week, the company introduced the MiMo-V2.5-Pro-UltraSpeed mode, achieving a record inference speed exceeding 1,000 tokens per second on general GPUs for trillion-parameter models. MiMo Code is built on the open-source OpenCode framework, featuring persistent memory, unlimited context, agent collaboration, and a proprietary Compose mode, while supporting major models such as DeepSeek, Kimi, and GLM under an MIT license.
MiMo Code offers optimized performance for code understanding, task planning, and tool invocation, reportedly outperforming frameworks like Claude Code. It also introduces voice input, enabling developers to issue spoken commands for coding tasks, aiming to improve collaboration efficiency and ease of use for both developers and general users.
