MiMo-V2-Pro
MiMo-V2-Pro is Xiaomi's text-only reasoning model built for the agent era, with over 1 trillion total parameters and 42 billion active per token. Released in March 2026, it was tuned for agentic workflows including complex tool calling, task orchestration, software development, and frontend web development inside agent frameworks. Its hybrid attention mechanism uses a 7:1 ratio of local to global attention, lifting capability over the V2-Flash variant while keeping inference efficient through a lightweight Multi-Token Prediction layer. With a 1 million token context window, it sustains high intensity agent flows involving thousands of sequential tool calls. MiMo-V2-Pro shows strong agentic performance, ranking number 3 globally on PinchBench at 81.0 and approaching Anthropic Claude Opus 4.6 on agent evaluation suites. Its coding ability surpasses Claude 4.6 Sonnet, with internal evaluations placing its code intelligence close to Claude Opus 4.6. Note: MiMo-V2-Pro has been superseded by MiMo-V2.5-Pro, which delivers further improvements in general agentic capability and complex software engineering.
Key info
Benchmarks
Independent benchmark scores — composite indices for reasoning, coding, and math, plus individual eval scores where available.
Available routes
No routes currently available — MiMo-V2-Pro isn't routed through the Opper gateway right now. It may return.
Contact us about this model →