MiniMax M2.7

Mar 2026
MiniMax M2.7
Compare in
on
—per task?
$0.30 / $1.20per 1M in / outcache $0.06
70 tok/sfirst token in 1.6 s?
205Kcontext

Rankingin thinking mode?

Strong at

S
Instructions#19 of 245
A
Reasoning#82 of 334Science#54 of 312Long context#61 of 273Knowledge#67 of 318
B
Agents & work tasks#78 of 175Tool calling#64 of 242
C
Coding#75 of 132

Alternatives?

ModelScoreCostPer taskWhat changes
CheaperGPT-5.4 Nanoxhighsame-12%$0.18BetterMiMo v2.5 ProthinkingOpen+5%+4%$0.054Memory (4-bit): 137 GB → 614 GBOpenMiniMax M3thinking+9%same$0.51Memory (4-bit): 137 GB → 256 GB+2CheaperGemma 4 31BthinkingOpensame-70%$0.10 / $0.34CheaperGemma 4 12BthinkingOpen-3%-71%$0.10 / $0.30

Run it yourself

estimate
Fits on✗ 8 GB (phone, laptop)✗ 16 GB✗ 24 GB GPU✗ 48 GB GPU✗ 80 GB GPU
Licencecustom licencecommercial use with conditions

History

6 versions
70758085Jan 26Apr 26Release dateIFBenchM2M2.1M2.5M2.7M3
70758085Jan 26Apr 26Release dateIFBenchM2M2.1M2.5M2.7M3

Benchmarks

all categories
Chat & Knowledge
Humanity's Last Exam29.6#82
AA Omniscience Index0.8#67
CritPt (physics)0.6#141
AA Intelligence Index 4.3.222.8#98
AA Omniscience · Accuracy26.8#98
AA Omniscience · Hallucination Rate35.6#51
AA-LCR (long context) 1.178.3#61
GDP.pdf · AA all-pass6#78
MLCR-AA3.3#53
GPQA Diamond87.4#54
Coding
Terminal-Bench 4.00#75
SciCode50.1#50
Terminal-Bench Hard (version unstated)39.4#34
SWE-bench Pro (edition unstated)lab56.2
Terminal-Bench 2.155.4#60
Agents & Tool Use
AutomationBench-AA4.4#88
GDPval-AA (work tasks) 2.125.5#78
τ³-Banking · AA9.9#90
AA Analyst Agent11.3#38
ITBench SRE26.5#32
APEX-Agents10.6#25
AA-Briefcase 1.1 · Elo718.6#83
τ²-bench · AA84.8#65
Writing & Language
IFBench (instructions)75.7#19

FAQ

How much does MiniMax M2.7 cost?›

$0.30 per 1M input tokens and $1.20 per 1M output tokens; cached input $0.06 per 1M; cheapest via DeepInfra at $0.25 / $1.

Is MiniMax M2.7 free to use?›

There is no free API; the cheapest paid price is $0.25 per 1M input tokens. The weights are open, so you can run it yourself.

How good is MiniMax M2.7, and what is it best at?›

#82 of 334 in Chat & Knowledge, #75 of 132 in Coding, #88 of 133 in Agents & Tool Use, #94 of 195 in Writing & Language, #78 of 133 in Vision & Documents. Strongest at: Reasoning, Science, Long context, Knowledge, Instructions.

Is MiniMax M2.7 better than MiniMax M2.5?›

On IFBench (instructions), MiniMax M2.7 scores 75.7 vs 71.6 — higher.

How can I use MiniMax M2.7?›

Through the API of AtlasCloud, DeepInfra, Groq, Minimax, Novita.

What is the context window of MiniMax M2.7?›

205K tokens, with up to 131K output tokens.

When was MiniMax M2.7 released? Is there a newer version?›

Released 2026-03-18. It has been superseded by MiniMax M3.

Is MiniMax M2.7 open source?›

Yes, the weights are open.

Can I run MiniMax M2.7 locally?›

The weights are open and can be self-hosted; hardware needs are not confirmed yet.

Can MiniMax M2.7 read images?›

No, it accepts text only.

What is MiniMax M2.7?›

MiniMax M2.7 is an open-weights language model from MiniMax, released Mar 2026, the successor to MiniMax M2.5.

What are the best alternatives to MiniMax M2.7?›

GPT-5.4 Nano (cheaper), MiMo v2.5 Pro (better), MiniMax M3 (open weights).