[Zhenzhuo HK Market Trends] Xiaomi (01810.HK) MiMo-V2.5 series large model officially begins public beta testing, MiMo-V2.5-Pro comprehensive intelligence index rated as globally leading
$XIAOMI-W(01810.HK) announced that its MiMo-V2.5 series of large models has officially entered public beta testing. According to the latest data from industry analysis firm Artificial Analysis, the Xiaomi MiMo-V2.5-Pro ranks joint first globally in the comprehensive intelligence index for open-source large models, placing it among the top five in the overall global large model rankings. Simultaneously, its Agent index also ranks joint first globally for open-source large models. This move signifies that domestic AI large models are accelerating their catch-up in core capabilities, even surpassing international leaders in some areas.
Flagship Model Focuses on Agent Capabilities, Will Be Open-Sourced
The MiMo-V2.5 series for this public beta includes the flagship MiMo-V2.5-Pro, the native multimodal large model MiMo-V2.5, as well as the MiMo-TTS and MiMo-ASR series models. Both the MiMo-V2.5-Pro and MiMo-V2.5 support million-level context windows and are planned to be open-sourced in the near future, which is expected to attract a large number of developers.
The flagship model MiMo-V2.5-Pro focuses on handling complex Agent tasks and can automatically execute multi-step instructions. In an official demonstration case, a compiler project that would have taken a Peking University graduate student several weeks to complete was finished by MiMo-V2.5-Pro in just 4.3 hours in a single attempt, receiving a perfect score. This capability demonstrates its powerful execution in a single task, capable of nearly a thousand tool calls and over a dozen context compressions.
Performance Rivals Top Models, Cost More Advantageous
In industry-standard benchmarks, the MiMo-V2.5 series has joined the leading echelon, competing head-to-head with top closed-source models like Claude Opus 4.6, Gemini 3.1 Pro, and GPT-5.4 in multiple evaluations. In the SWE-bench Pro test, which most closely resembles real-world software development scenarios, the performance of MiMo-V2.5-Pro is already on par with Claude Opus 4.6 and GPT-5.4.
Beyond performance, Xiaomi also emphasizes its cost-effectiveness. While achieving the same Agent evaluation scores, MiMo-V2.5-Pro saves a quarter of the Token consumption compared to Kimi K2.6; the MiMo-V2.5 model can save nearly double the Tokens compared to Claude Sonnet 4.6. For enterprise users, this means lower operational costs.
Eliminates Long Context Premium: More Performance, Lower Pricing
To accompany the public beta, Xiaomi simultaneously updated its Token Plan billing system, removing the differential Credit multiplier for 1M and 256k context windows, and introduced multiple discounts including 20% off for nighttime usage, 30% off for auto-renewal, and 12% off for annual subscriptions. Furthermore, all Credits used by existing Token Plan subscribers up to 10 PM on April 22nd will be reset to zero to align with the new billing system.
The copyright of this article belongs to the original author/organization.
The views expressed herein are solely those of the author and do not reflect the stance of the platform. The content is intended for investment reference purposes only and shall not be considered as investment advice. Please contact us if you have any questions or suggestions regarding the content services provided by the platform.
