MiniMax M2

Item: MiniMax M2
Rating: 2
Author: Ali Madjaji

MiniMax · China

229B agent-focused model from MiniMax, Modified MIT. Strong software-engineering and tool-use benchmarks. Family has iterated fast (M2 / M2.1 / M2.5 / M2.7 across 2025-2026). Same Chinese-origin alignment and supply-chain considerations as DeepSeek, Qwen, Kimi.

Sovereignty

Licence: Modified MITCommercial: YesTraining data: UndisclosedOrigin: China

Licence facts

Licence: Modified MIT
Commercial use: Unrestricted
Derivatives: Allowed
Attribution: Minimal
Parameters: 229B
Architecture: Lightning Attention (per MiniMax-01 paper)
Training data: Not disclosed
Last updated: 2025-Q4 (M2); M2.7 most recent variant

Performance & pricing?

Quality index36/100Speed72 tok/sBlended price$0.53/M

Known risks

Chinese-origin alignment biases
Rapid versioning (M2→M2.7 in 4 months) makes pinning risky
Modified MIT — verify the specific modifications

Sources

→ Model card → Latest variant

Compare with

vsStep-3.7-Flash vsHy-MT2-30B-A3B vsMiniCPM5-1B vsMixtral 8x22B Instruct

Reviewed by Ali Madjaji · Last reviewed 2026-04-16· Reviewed 67 days agoSuggest a correction