Minimax M2.5
Parameters
-
Context Length
128K
Modality
Multimodal
Architecture
Dense
License
Proprietary
Release Date
15 Dec 2025
Knowledge Cutoff
-
Evaluation Benchmarks
Rank
#74
| Benchmark | Score | Rank |
|---|---|---|
Agentic Coding LiveBench Agentic | 0.52 | 17 |
StackUnseen ProLLM Stack Unseen | 0.66 | 17 |
Graduate-Level QA GPQA | 0.81 | 22 |
Mathematics LiveBench Mathematics | 0.77 | 27 |
Professional Knowledge MMLU Pro | 0.80 | 35 |
Coding LiveBench Coding | 0.71 | 36 |
Data Analysis LiveBench Data Analysis | 0.50 | 38 |
Reasoning LiveBench Reasoning | 0.59 | 40 |
Web Development WebDev Arena | 1382 | 47 |
General Text Text Arena | 1390 | 57 |
Rankings
Overall Rank
#74
Coding Rank
#62
About Minimax M2.5
Minimax M2.5 is an advanced multimodal model offering state-of-the-art text generation and reasoning capabilities. Features strong multilingual support with particular emphasis on Chinese and English. Designed for versatile applications including content generation, analysis, and conversational AI with competitive performance across multiple benchmarks.
Technical Specifications
Attention
Attention Structure
Multi-Head Attention
Attention Heads
-
Key-Value Heads
-
Attention Head Dimension
-
Position Embedding
Absolute Position Embedding
RoPE Theta
-
Sliding Window Attention
-
Sliding Window Size
-
Sliding Window Ratio
-
Linear Attention
-
Linear Attention Ratio
-
Normalization
-
Activation Function
-
Dimensions
Hidden Dimension Size
-
Number of Layers
-
FFN Intermediate Size (Dense)
-
Multi-Token Prediction Heads
-
Tokenizer
Vocabulary Size
-
Resources
About Minimax M2
Minimax M2.5 series represents cutting-edge multimodal AI models from Minimax AI, featuring state-of-the-art performance in text generation, reasoning, and multilingual understanding. These models combine high-quality language understanding with efficient architecture, optimized for both API deployment and enterprise solutions.
Other Minimax M2 Models
- No related models available