Xiaomi unveils the 'MiMo-V2.5-Pro,' an AI model that rivals the Claude Opus 4.6, as an open model.

Xiaomi has unveiled its AI models, the ' MiMo-V2.5 ' and ' MiMo-V2.5-Pro ,' as open models. The MiMo-V2.5-Pro is being promoted as a model that surpasses the Gemini 3.1 Pro and approaches the performance of the Claude Opus 4.6.
MiMo-V2.5 | Xiaomi
MiMo-V2.5-Pro | Xiaomi
https://mimo.xiaomi.com/mimo-v2-5-pro
MiMo-V2.5 is a MoE model with a total of 310 billion parameters and 15 billion active parameters, while MiMo-V2.5-Pro is a MoE model with a total of 1.02 trillion parameters and 42 billion active parameters.
The benchmark results for 'MiMo-V2.5-Pro', 'MiMo-V2.5', 'MiMo-V2-Pro', 'Claude Opus 4.6', 'Gemini 3.1 Pro', and 'GPT-5.4' are shown below. The MiMo-V2.5-Pro recorded higher scores than the Gemini 3.1 Pro in most tests.

In Xiaomi's in-house benchmark, MiMo Coding Bench, the MiMo-V2.5-Pro beat the Gemini 3.1 Pro and demonstrated performance close to that of the Claude Opus 4.6. The MiMo Coding Bench is a test that measures coding capabilities when using agent frameworks such as Claude Code, and based on these results, Xiaomi is promoting the MiMo-V2.5-Pro as a 'low-cost model with top-class intelligence.'

MiMo-V2.5-Pro is also capable of autonomously executing lengthy tasks. In a compiler development task based on

The third-party organization

In terms of coding performance, it scored lower than DeepSeek V4.

In terms of agent performance, it demonstrates performance comparable to closed-source frontier models.

MiMo-V2.5-Pro and MiMo-V2.5 are available on Hugging Face. They are licensed under the MIT License.
XiaomiMiMo/MiMo-V2.5 · Hugging Face
https://huggingface.co/XiaomiMiMo/MiMo-V2.5

XiaomiMiMo/MiMo-V2.5 · Hugging Face
https://huggingface.co/XiaomiMiMo/MiMo-V2.5

Related Posts:
in AI, Posted by log1o_hf







