The Cost-Effectiveness of MiniMax's Open-Source M2 Model in Agent and Coding Scenarios
2025-10-27 / Read about 0 minute
Author:小编   

On October 27, 2025, MiniMax made a groundbreaking announcement—the open-sourcing of its M2 model. This model has been meticulously crafted for Agent and coding applications. Leveraging a Mixture of Experts (MoE) architecture, it boasts an impressive total of 230 billion parameters, with 10 billion being actively utilized. Such a design enables it to outperform even the most elite models, such as Gemini 2.5 Pro, in terms of performance.