AI News (2026/2/13): MiniMax M2.5 Released: $1/Hour, King of Real-World Work
Executive Summary:
MiniMax has officially released the M2.5 model, which achieves or sets new industry SOTA levels in advanced productivity scenarios such as programming, search tool invocation, and office work. The M2.5 model significantly improves task decomposition and reasoning efficiency, completing tasks 37% faster on the SWE-Bench Verified programming benchmark compared to the previous M2.1 model.
News Details
MiniMax officially released its latest multimodal large model, M2.5, on February 13. This model excels in advanced productivity scenarios such as programming, search tool invocation, and office work, achieving or setting new industry SOTA levels. The core improvements of M2.5 lie in enhanced task decomposition and reasoning efficiency, making it more efficient and reliable in practical applications.
Key Points
Task Decomposition: The M2.5 model demonstrates outstanding task decomposition capabilities, breaking down complex tasks into multiple subtasks with greater precision, thereby improving execution efficiency and accuracy. For example, on the programming benchmark SWE-Bench Verified, M2.5 completes tasks 37% faster than the previous M2.1 model.
Reasoning Efficiency: M2.5 also shows significant improvements in reasoning efficiency, especially when handling multimodal data. The model can quickly understand and generate various types of data, including text and images, making it more efficient in applications such as RAG (Retrieval-Augmented Generation).
Cost and Availability: MiniMax announced that the usage cost of the M2.5 model is $1/hour, making high-performance AI tools more affordable for more developers and enterprises. Additionally, M2.5 has been made available to some developers for testing and is planned to be fully released in the coming months.
AI-ALL In-Depth Review
The release of MiniMax M2.5 marks a significant step forward in the maturity of multimodal large models in advanced productivity scenarios. The notable improvements in task decomposition and reasoning efficiency not only resulted in excellent performance on the SWE-Bench Verified programming benchmark but also demonstrated strong potential in real-world applications. The usage cost of $1/hour makes high-performance AI tools more accessible, promoting the widespread adoption of AI technology in areas such as development and office work. However, the full release of M2.5 will take some time, during which MiniMax can gather further feedback and optimize model performance to ensure stability and reliability in large-scale applications.
