MiniMax to open-source 2.7T-parameter M3 Pro, China's largest

MiniMax, a Chinese AI startup known for its conversational and multimodal models, is reportedly preparing to open-source its largest model yet: M3 Pro, a 2.7-trillion-parameter beast. If released as planned in the third quarter of 2025, it would become China's largest open-source AI model, outpacing other domestic efforts like Alibaba's Qwen and Baidu's ERNIE. The move signals a strategic bet on openness to attract global developer mindshare and challenge Western open-source models such as Meta's Llama 3.1 (405B) and Mistral's large offerings. With 2.7 trillion parameters, M3 Pro would require significant computational resources to run, but MiniMax is likely optimizing it for efficiency and distributed inference.

This open-source release could accelerate AI development in China by giving researchers and startups access to a state-of-the-art foundation model without proprietary restrictions. It also intensifies the AI arms race between the US and China, as open-source models lower barriers to entry for advanced capabilities. MiniMax has previously demonstrated strong performance with its MiniMax-01 and MiniMax-Text-01 models, and M3 Pro is expected to push boundaries in reasoning, multilingual understanding, and long-context processing. Developers should watch for benchmark results and licensing details, which will determine how broadly the model can be used commercially. If successful, M3 Pro could become a cornerstone of China's open-source AI ecosystem and a credible alternative to Western models.