Xiaomi's MiMo-V2-Pro Cracks the Global Top 10 for AI Reasoning — at a Fraction of the Price

Xiaomi has quietly pushed its way into the global front rank of AI reasoning. The company's MiMo-V2-Pro scored 49 on the independent Artificial Analysis Intelligence Index, landing the model at #10 in the world — wedged between Kimi K2.5 and GLM-5, and level with OpenAI's GPT-5.2 Codex tier — while charging a small fraction of what comparable Western frontier models cost to run.

MiMo-V2-Pro is a dedicated reasoning model, using extended chain-of-thought to work through problems before answering, and it ships with a 1-million-token context window. It is the successor to Xiaomi's open-weights MiMo-V2-Flash (309B total / 15B active parameters, MIT-licensed), but this Pro release is, for now, available only through Xiaomi's first-party API — the weights have not been published.

The benchmark detail is where it gets interesting. On the GDPval-AA economic-task evaluation, MiMo-V2-Pro posted an Elo of 1426, ahead of GLM-5 (1406) and well clear of Kimi K2.5 (1283). Its hallucination rate on the AA-Omniscience test fell to roughly 30%, a sharp improvement on the Flash model's 48%. Crucially, it is far more token-efficient than its rivals, spending about 77 million output tokens to complete the full index versus GLM-5's 109 million.

That efficiency translates straight into price. MiMo-V2-Pro is listed at $1 per million input tokens and $3 per million output (at 256K context), rising to $2 / $6 for the full 1M-token window. Artificial Analysis pegged the total cost of running its entire Intelligence Index suite at just $348 — a number that undercuts most models in the same intelligence tier and underlines why low-cost Chinese reasoning models keep narrowing the gap with the labs in San Francisco.

The reasoning win is a different story from the throughput record Xiaomi set earlier this month, when its MiMo-V2.5-Pro-UltraSpeed variant became the first trillion-parameter model to clear 1,000 tokens per second on commodity hardware. Taken together, the two releases sketch a company attacking the frontier from both ends at once — raw intelligence and raw speed — and doing it at prices that are increasingly hard for anyone to ignore.