Quick Facts

  • MiMo-V2-Pro scored 61.5 on ClawEval benchmark, approaching Claude Opus 4.6’s 66.3 score and beating GPT-5.2’s 50.0
  • Model costs $348 to run full benchmark tests compared to $2,304 for GPT-5.2 and $2,486 for Claude Opus 4.6
  • Features 1T total parameters with 42B active and supports up to 1M-token context

Xiaomi released its MiMo-V2-Pro large language model on March 19, achieving performance levels that approach top-tier AI models at a fraction of the operating cost. Third-party benchmarking organization Artificial Analysis verified the claims, ranking MiMo-V2-Pro at #10 globally on its Intelligence Index.

The model scored 61.5 on ClawEval, a benchmark for agentic scaffolds, coming close to Claude Opus 4.6’s score of 66.3 while significantly outpacing GPT-5.2’s 50.0. On coding-specific tests like Terminal-Bench 2.0, MiMo-V2-Pro achieved an 86.7 score.

Cost efficiency sets MiMo-V2-Pro apart from competitors. Running the complete Artificial Analysis Intelligence Index required only 77M output tokens, compared to 109M for GLM-5 and 89M for Kimi K2.5. The model’s pricing starts at $1 per million tokens for input within smaller contexts.

“Xiaomi’s actual progress in AI might be much faster than what everyone sees,” CEO Lei Jun said on Weibo. The company expects AI R&D and capital investment to exceed 16 billion yuan this year.

Goldman Sachs upgraded its outlook on Xiaomi following the release, maintaining a “Buy” rating with a target price increase to HKD 41. The investment bank stated the concentrated release of three flagship models marks Xiaomi’s transition from AI R&D investment to achievement realization.

The model uses a hybrid architecture with 1T total parameters and 42B active parameters, roughly three times larger than MiMo-V2-Flash. It supports up to 1M-token context and ranks 8th worldwide among all LLMs and 2nd among Chinese models.

Before Xiaomi revealed its identity, the model operated under the codename “Hunter Alpha” during testing. It accumulated over one trillion tokens of usage on OpenRouter, topping daily API call volume lists for multiple days.

Xiaomi integrated all three models into WPS Office, Xiaomi phones and computers, and the company’s browser. The models are also partnering with five major agent development frameworks including OpenClaw and OpenCode to offer one week of free API access for developers worldwide.

Read more: Xiaomi stuns with new MiMo-V2-Pro LLM nearing GPT-5.2, Opus 4.6 performance at a fraction of the cost

This article was written by an AI agent. Spotted an error? Send a correction and we will fix it.