MiniMax-M2.5

Parameters

229B

Published

Feb 12, 2026

MiniMax-M2.5 is an AI model developed by MiniMax (China), first published in February 2026. It works in the language domain, on tasks such as code generation, system control, search and 2 more.

Epoch AI has no training-compute estimate for this model. The model has 229,000,000,000 parameters.

Access: Open weights (unrestricted). Its weights are openly available. Epoch AI rates the confidence of this record as likely.

Full record

Organization: MiniMax
Country of organization: China
Domain: Language
Task: Code generation, System control, Search, Language modeling/generation, Question answering
Parameters: 229,000,000,000
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Epoch confidence: Likely

Benchmark results

01	WebDev Arena	Arena Score	1382.1
02	ALE Bench	Performance	618.2
03	Epoch Capabilities Index	ECI Score	147.1
04	ARC-AGI	Score	0.64
05	Terminal-Bench	Accuracy mean	0.43
06	CL Bench	Overall	0.11
07	Posttrainbench	Average (%)	0.1
08	CL Bench Life	Overall	0.06
09	Apex Agents	Pass@1 score	0.06
10	ARC-AGI-2	Score	0.05
11	Proofbench	Accuracy	0.04
12	Vending-Bench 2	Score	-23.16