Open source Xiaomi MiMo-V2.5 and V2.5-Pro are among the most efficient (and affordable) at agentic 'claw' tasks

· ai ai-agents open-source · Source ↗

TLDR

  • Xiaomi releases MiMo-V2.5 and MiMo-V2.5-Pro under MIT License, claiming top efficiency on agentic claw benchmarks at competitive prices.

Key Facts

  • MiMo-V2.5-Pro leads open-source models on ClawEval with a 63.8% success rate, using roughly 40-60% fewer tokens than Claude Opus 4.6, Gemini 3.1 Pro, and GPT-5.4.
  • The Pro model is a 1.02 trillion-parameter MoE with 42B active parameters; the base V2.5 is 310B total with 15B active.
  • API pricing starts at $0.40/M input tokens for MiMo-V2.5 and $1.00/M for MiMo-V2.5-Pro (standard context), with cache hits as low as $0.20/M.
  • Full MIT License allows commercial deployment, fine-tuning on proprietary data, and distribution of derivative weights without Xiaomi’s permission.

Why It Matters

  • MIT licensing with no usage caps lowers the barrier for enterprises to self-host and build on these models without ongoing compliance risk from custom open-weight licenses.
  • Token efficiency gains are directly relevant as more services shift to usage-based billing, making per-trajectory cost a real differentiator for agentic workloads.

Carl Franzen / VentureBeat · 2026-04-27 · Read the original

Related coverage