Xiaomi's agentic AI coding harness MiMo Code beats Claude Code at 200 step tasks (venturebeat.com)

Xiaomi has launched MiMo Code V0.1.0, an open-source AI coding assistant that reportedly outperforms Anthropic's Claude Code in handling complex, long-horizon coding tasks. According to Xiaomi's internal tests involving 576 developers, MiMo Code excelled particularly on tasks exceeding 200 execution steps, achieving a win rate of over 65% against Claude Code. This performance boost is attributed to its innovative memory architecture, which employs a cross-session memory system utilizing SQLite FTS5 for efficient information retrieval and persistent storage, allowing the AI to maintain context without losing operational momentum during extensive coding sessions. The MiMo Code is designed for seamless integration into developers' existing workflows, operating directly within terminal environments and featuring voice control capabilities. Its bundled model, MiMo-V2.5, offers a significant context window of one million tokens, requiring zero configuration to start. With competitive pricing and ease of use, Xiaomi aims to disrupt the terminal-based AI coding market, which is dominated by players like OpenAI, Anthropic, and Google. By offering a robust tool that emphasizes memory management and developer collaboration, Xiaomi is positioning itself as a formidable contender in the rapidly evolving landscape of AI-powered coding solutions.