Xiaomi’s MiMo Code V0.1.0 Open Source Release
Xiaomi’s MiMo AI team open-sourced MiMo Code V0.1.0 on June 10, 2026 — a terminal-native AI coding assistant the company says outperforms Anthropic’s Claude Code on long-horizon multi-step tasks (200+ steps), based on internal beta and a survey of 576 developers. Benchmark claims are vendor self-reported and not independently verified.
Memory Architecture
Cross-session memory via SQLite FTS5 spanning four layers: project memory (MEMORY.md), session checkpoints, scratch notes, and per-task progress logs. An independent checkpoint-writer subagent updates state while the main agent continues working.
Self-improvement: /dream (weekly session review and compression) and distill (workflow automation from repeated patterns).
Benchmark Claims (Xiaomi Internal)
MiMo Code + MiMo-V2.5-Pro vs Claude Code + Claude Sonnet 4.6:
| Benchmark | MiMo Code | Claude Code |
|---|---|---|
| SWE-bench Verified | 82% | 79% |
| SWE-bench Pro | 62% | 55% |
| Terminal Bench 2 | 73% | 69% |
Harness-only gain (same MiMo-V2.5-Pro model): +5 points on SWE-bench Pro and Terminal Bench 2.
Past 200 execution steps, MiMo Code win rate rose above 65% in human A/B testing (1,213 judged pairs).
MiMo-V2.5 Model
Sparse mixture-of-experts: 310B total parameters, 15B active per inference, 1M token context. MiMo Auto channel offers free limited-time access with zero configuration.
Pricing: MiMo-V2.5 from 2.00/M output tokens.
Enterprise Considerations
- MIT license permits modification and commercial integration
- OpenCode lineage makes architecture inspectable
- BYOM support for internal endpoints
- “Free for limited time” MiMo Auto routes code context through Xiaomi servers — concern for IP-sensitive organizations
- V0.1.0 signals early maturity
- U.S. government procurement restrictions on Chinese vendors may apply
Competitive Landscape
Terminal agents: Claude Code, OpenAI Codex CLI, Google Gemini CLI, OpenCode, Aider. Xiaomi follows DeepSeek/Qwen/Kimi pattern: capable models and tooling under permissive licenses at fraction of U.S. lab pricing.