Artificial Analysis@ArtificialAnlys

2026-04-24 14:19·69天前

AI 摘要

小米于2026年4月22日发布MiMo V2.5 Pro模型，其在Artificial Analysis智能指数得分54，表现强劲。该模型在代理能力上领先，Agentic Index得分67%，优于Kimi K2.6等开源模型，这主要得益于在τ²-Bench和TerminalBench Hard上的优异表现。此外，其在前沿推理和指令遵循方面也有提升。模型具有1M上下文窗口，1T总参数，API定价具有成本效益，运行同一智能指数仅需462美元。不过，其在事实准确性指数上得分4，存在准确率较低的问题。模型采用MIT许可证，权重暂未公开，目前可通过小米官方API使用。

Xiaomi has just released MiMo V2.5 Pro which performed very strongly on Artificial Analysis Intelligence Index （54）. It is one of the leading models on agentic index and is expected to open source its weights.

@Xiaomi's MiMo V2.5 Pro shows an impressive improvement over MiMo V2 Pro （49）， the previous generation of Xiaomi's flagship model family， which was released just over a month ago on March 19， 2026. MiMo V2.5 Pro will be open weights， but weights have not yet been released. With continued progress and an accelerated release cadence， Xiaomi， currently trails only OpenAI， Anthropic， Google， and Moonshot in intelligence

Key takeaways：

➤ MiMo V2.5 Pro leads open weights models in agentic capabilities with a 67% Agentic Index score. This is higher than both Kimi K2.6 （66%） and MiniMax M2.7 （61%）， driven by τ2-Bench （94%） and strong TerminalBench Hard performance （43%）. This indicates the model would be more capable in agentic reasoning workflows.

➤ Other gains are made in frontier reasoning and instruction following. The model scores 34% on HLE （+6% from MiMo V2.0） and 80% on IFBench （+11% from MiMo V2.0）. However， compared to the previous generation， we saw a small regression in CritPt （5% to 4%）.

➤ MiMo V2.5 Pro's token efficiency remains competitive against peers in a similar intelligence tier， using ~92M output tokens for the Intelligence Index. This is more efficient than Kimi K2.6 （~170M） and GLM 5.1 （~110M）. However， it does use 19% more than the previous generation MiMo V2 Pro （77M）.

➤ Priced at $1.00/$3.00 per M input/output tokens on Xiaomi's First Party API， MiMo V2.5 Pro is relatively cost-efficient for its intelligence tier. It costs only $462 to run the Artificial Analysis Intelligence Index， compared to $948 for Kimi K2.6 and $544 for GLM 5.1.

➤ MiMo V2.5 Pro scores 4 on the AA-Omniscience Index， a proprietary Artificial Analysis evaluation that measures factual accuracy and hallucinations. This is a slight regression from MiMo V2 Pro （5）， though both models still trail proprietary frontier models. MiMo V2.5 demonstrates a relatively low hallucination rate （25%） but also low accuracy （23%）.

Artificial Analysis@ArtificialAnlys · X

63导出 Markdown

2026-04-24 14:19·69天前

在 X 看原推· x.com

AI 摘要