Chinese Open-Source Coding Agents Are Catching Up Fast

Xiaomi released MiMo Code, an open-source terminal-native AI coding assistant that can read and write code, run commands, manage Git, and use a browser. It is built as a fork of OpenCode, which immediately drew a pointed comment: 'Why not just contribute to OpenCode instead of creating a clone?' Separately, Moonshot AI dropped Kimi K2.7-Code, an open-source coding model with claimed better token efficiency than existing options, and the thread quickly turned into a price-per-token comparison with Anthropic's Opus, which is roughly 5-7x more expensive than Chinese alternatives at comparable capability levels.

The pattern: the gap between frontier Western closed models and open-weight Chinese models is closing on coding benchmarks specifically. Commenters in the Kimi thread noted that any new model not demonstrably 20-30% better than DeepSeek v4 and priced above DeepSeek's token cost is 'automatically deprecated.' That is a brutal framing but it reflects real market pressure. Xiaomi shipping a coding agent, Moonshot shipping a coding model, and DeepSeek already running in many production stacks means the open-weight coding tier is now genuinely competitive.

The counterpoint is that benchmark performance on coding tasks does not always translate to production reliability, and several commenters expressed interest in real-world setup stories from people using open-weight models full-time before making a switch.