Creator

Xiaomi

Released

2026-04-22

Intelligence

37.2

Artificial Analysis Index

Coding

56.8

Artificial Analysis Index

In $/1M

$0.14

input tokens

Out $/1M

$0.28

output tokens

Blended $/1M

$0.17

3:1 blended

Speed

81

tokens / sec

Profile

License, openness, and modality — the governance layer.

License

MIT

commercial OK

Weights

Open

downloadable

Modalities

Audio · Image · Text · Video

Parameters

311B

total

MiMo-V2.5 is Xiaomi's native omnimodal sparse Mixture-of-Experts model with 310B total parameters, 15B activated parameters, and a 1M-token context window. Built on the MiMo-V2-Flash backbone, it adds dedicated vision and audio encoders for text, image, video, and audio understanding, and is post-trained with SFT, agentic reinforcement learning, and Multi-Teacher On-Policy Distillation for multimodal perception, long-context reasoning, and agentic workflows.

Capability profile

Category strength across 20 domains, via LLM Stats.

long context
90vision
80multimodal
80general
70reasoning
70
tool calling
70code
60video
60agents
60finance
40
Via LLM Stats

Benchmark breakdown

Independent evaluation scores, normalized to 0–100.

GPQA Diamond
85
Humanity’s Last Exam
25
MMLU-Pro
SciCode
43
LiveCodeBench
MATH-500
AIME 2025
τ²-Bench (agentic)
91
Terminal-Bench Hard
42
IFBench
67
Via Artificial Analysis

Providers

Inference hosts serving this model — their own pricing and measured performance, cheapest input first.

ProviderIn $/1MOut $/1MThroughputLatencyStatus
Novita$0.17$0.34active
DeepInfra$0.40$2.00active
All providers

2 hosts · via LLM Stats

© 2026 NYSGPT2525 LLC