Model Comparison
MiniMax-M2
vs
Qwen2.5-VL-72B-Instruct
Feb 28, 2026
Pricing
Input
$
0.3
/ M Tokens
$
0.59
/ M Tokens
Output
$
1.2
/ M Tokens
$
0.59
/ M Tokens
Metadata
Create on
Oct 22, 2025
Jan 27, 2025
License
MIT
-
Provider
MiniMaxAI
Qwen
Specification
State
Deprecated
Available
Architecture
Mixture of Experts
Vision-Language Model (VLM) with a Streamlined and Efficient Vision Encoder (ViT with window attention, SwiGLU, RMSNorm) aligned with the Qwen2.5 LLM structure. Features include Dynamic Resolution and Frame Rate Training for video understanding, mRoPE for temporal sequence and speed, and YaRN for long text context length extrapolation.
Calibrated
No
No
Mixture of Experts
Yes
No
Total Parameters
230B
72B
Activated Parameters
10B
72B
Reasoning
No
No
Precision
FP8
FP8
Context length
197K
131K
Max Tokens
131K
4K
Supported Functionality
Serverless
Supported
Supported
Serverless LoRA
Not supported
Not supported
Fine-tuning
Not supported
Not supported
Embeddings
Not supported
Not supported
Rerankers
Not supported
Not supported
Support image input
Not supported
Not supported
JSON Mode
Supported
Not supported
Structured Outputs
Not supported
Not supported
Tools
Supported
Not supported
Fim Completion
Not supported
Not supported
Chat Prefix Completion
Supported
Supported
MiniMax-M2 in Comparison
See how MiniMax-M2 compares with other popular models across key dimensions.
VS
MiniMax-M2.5
VS
Step-3.5-Flash
VS
MiniMax-M2.1
VS
Qwen3-VL-235B-A22B-Instruct
VS
Qwen3-VL-235B-A22B-Thinking
VS
Ring-flash-2.0
VS
Qwen3-Next-80B-A3B-Instruct
VS
Qwen3-Next-80B-A3B-Thinking
VS
gpt-oss-120b
VS
step3
VS
Qwen3-235B-A22B-Thinking-2507
VS
Qwen3-Coder-480B-A35B-Instruct
VS
Qwen3-235B-A22B-Instruct-2507
VS
Qwen2.5-VL-72B-Instruct
VS
Qwen2.5-72B-Instruct
VS
Qwen2.5-72B-Instruct-128K
VS
VS
