MiniMax M2

MiniMax llm

A 230B-parameter (10B active) MoE model that deliberately returned to full attention over M1's linear attention, optimized for coding and agentic workflows.

overall ⌄

35.0

101 of 125 ranked

price · $/M tokens

0.3 / 1.2

in / out, as of 2026-07-25

to run it yourself ⌄

138 GB

minimum · 552 GB recommended

Benchmarks

What its maker published, and what anyone else measured. Every figure links the document it came from.

benchmarkpublishedmeasured
MMLU-Pro82 —
GPQA78 —
AIME 202578 —
SWE-bench69.4 61
HLE12.5 —
BrowseComp44 —

Cheaper, and at least as good

On a 1,000-in, 1,000-out call these cost less and rank no lower — a hard comparison to argue with, and a narrow one. Value covers what it misses.

Devstral Small 2 Mistral AI$0.00040overall 44.3
Ministral 3 14B Reasoning Mistral AI$0.00040overall 47.5
DeepSeek V4-Flash DeepSeek$0.00042overall 57.9
Claude Haiku 5.5 Anthropic$0.00060overall 65.3
Qwen3.8-Flash-Next Alibaba$0.00062overall 61.2

Lineage

↓ finetune MiniMax M2.1
view in full graph →

Metadata

orgMiniMax
released2025-10-27
params230B-A10B
licensemodified-mit
hf idMiniMaxAI/MiniMax-M2