MiMo-V2.5-Pro

First open-weight MiMo Pro. Superseded by MiMo-V2.6-Pro; API name retires Oct 21, 2026.

Released

Apr 22, 2026

Price in / out

—

Blended $/1M

—

Context

Parameters: 1T / 42B active. License: MIT.

Benchmark scoresvs MiMo-V2.5

Benchmark	Score	Δ vs prior	Rank (current)	Configuration	Source
Terminal-Bench 2.1	57.3%	-3.4	#19 of 23	Vals AI run · independent run	vals.ai ↗
GPQA Diamond	82.6%	+1.0	#19 of 22	Vals AI run · independent run	vals.ai ↗
Humanity's Last Exam	34%	—	#25 of 32	no tools; Xiaomi reports 48.0 with tools · lab-reported	mimo.xiaomi.com ↗
τ²-bench (Telecom)	94.2%	—	#4 of 10	Artificial Analysis telecom run · independent run	artificialanalysis.ai ↗
MMLU-Pro	84.6%	+1.7	#10 of 13	Vals AI run · independent run	vals.ai ↗

MiMo versions

Timeline

Version	Released	$ in / out	Terminal-Bench	SWE-bench	GPQA	HLE	τ²-bench	MMLU-Pro
MiMo-V2-FlashSuperseded	2025-12-16	$0.1 / $0.3	—	73.4%	83.7%	—	83.9%	84.9%
MiMo-V2-ProRetired	2026-03-18	$1 / $3	—	78%	87%	—	95%	—
MiMo-V2.5Superseded	2026-04-22	—	60.7%	—	81.6%	—	—	82.9%
MiMo-V2.5-ProSuperseded	2026-04-22	—	57.3%	—	82.6%	34%	94.2%	84.6%
MiMo-V2.6-Flash	2026-09-21	$0.14 / $0.28	76.4%	—	—	35.1%	—	—
MiMo-V2.6-Pro	2026-09-21	$0.435 / $0.87	67.8%	—	—	49.4%	—	—

Compare

Open the Comparison Lab with MiMo-V2.5-Pro and its open-weight peers pre-selected, then toggle any benchmark or model.

Every MiMo version over time

Shipping with it?

Benchmarks are a starting point. If you run MiMo-V2.5-Pro in production, post what it does well and where it falls over; other founders are deciding right now.

Write about MiMo-V2.5-Pro