MiniMax is preparing a next-generation large language model codenamed M3 Pro, which will feature 2.7 trillion parameters. The company intends to release the model and open-source it as early as the third quarter of 2025.

This new version represents a massive scale increase over the current flagship M3 model, which has 428 billion parameters. By significantly expanding the parameter count, MiniMax aims to enhance the AI's ability to handle multi-step instructions and perform complex reasoning tasks.