MiniMax has directly surged to 2.7 trillion parameters. If it truly goes open-source, the landscape of domestic large models will be upended.

CoinWorld news, MiniMax is developing a large language model with 2.7 trillion parameters, which may be released as early as the third quarter, with plans to open-source it. The internal codename for the model is m3 pro, but the final release name has not been determined. If the weights are open-sourced as planned, this model could become the largest open-source large model in the world by parameter count. MiniMax's current flagship model, m3, has 428 billion parameters, and the new model's scale is approximately 6.3 times that of m3, also surpassing several existing trillion-parameter open-weight models. A larger parameter count generally benefits complex reasoning and multi-step instruction tasks, but final performance still depends on training quality, activated parameters, inference cost, and benchmark results.