Xiaomi MiMo-V2-Pro: #3 Globally on Agent Benchmarks (ClawEval 61.5) at 20% of Claude Opus 4.6 Cost

VentureBeat·high signal

Xiaomi launched MiMo-V2-Pro on March 18 — a 1T-parameter model purpose-built for agentic workloads. It scores 61.5 on ClawEval (Claude Opus 4.6 scores 66.3), ranking it third globally on agent tasks at roughly 20% of the API price. Before the official launch, it ran as 'Hunter Alpha' on OpenRouter in stealth, processing over 1 trillion tokens before its origin was identified. MiMo-V2-Flash (309B, open source) also beats most models at its weight class.