ChainThink report, on July 8, according to sources cited by The Information, MiniMax is developing a large language model with 2.7 trillion parameters, potentially launching as early as the third quarter and planning to open-source its weights.

The model's internal codename is M3 Pro, with the final public name yet to be determined. If released with open weights as planned, it could become the largest open-source LLM globally in terms of parameter count.

MiniMax’s current flagship model, M3, has 428 billion parameters; the new model will scale approximately 6.3 times larger, surpassing several existing trillion-parameter open-weight models.

The report also notes that larger parameter counts generally favor complex reasoning and multi-step instruction tasks, but final model performance still depends on training quality, activated parameters, inference cost, and benchmark results.