Xiaomis MiMo Code claims it beats Claude Code past 200 steps

▤ Summary

Xiaomi’s MiMo Code is an open-source coding agent built for terminal-native, long-horizon software tasks. The article centers on the company’s claim that it can outperform Claude Code on runs that exceed 200 steps, though the evidence cited is self-reported. It also frames endurance in agentic coding as a procurement issue for enterprises that need reliable delivery, checkpoints, and resumable workflows. The piece compares MiMo Code with other approaches such as Anthropic’s subagents, Arbor’s persistent hypothesis tree, and Berkeley’s Agents Last Exam. The main takeaway is that buyers should evaluate coding agents on how long they stay coherent in real work, not just on benchmark headlines.

▦ Linked Companies

Anthropic

$10M to $25M