Xiaomimodel systemcard
MiMo V2.5 Pro
#16of 27 · general
non-reasoning mode collapsed; the plain V2.5 and the older V2 Pro are different models and stay off the board
Standing
dot = this model · shaded = the field's middle 60% · track = full fieldGeneral
29.0#16/27
Math
32.1#18/27
Coding
40.4#14/27
Health
9.7#23/26
Engineering
12.5#15/17
Analyst
30.0#8/11
Legal
50.9#11/26
Instruct
61.0#8/27
Agents
24.2#21/27
Professional
43.9#14/27
Hallucination
39.4#11/27
Speed
32.6#16/22
Value
41.1#16/27
Open
29.6#7/11
Measured
≈ imputed · amber dot = lab-self-reported among the sourcesGPQA Diamond82.6%#24
HLE34.8%#20
MMLU-Pro76.5%#25
SWE-bench V76.5%#22
LiveCodeBench81.3%#20
SciCode50.2%#17
CritPt4.0%#21
Terminal-Bench65.2%#17
APEX Agents2.4%#19
GDPval38.3%#16
Arena1465#13
AA Index42.9#18
Vals Index50.7#19
Operations
$0.43 / $0.87$ per 1M tokens, in / out
$0.54blended
62tokens / sec
1465arena elo · 43,790 votes