Chinese AI developer MiniMax is building a massive large language model with 2.7 trillion parameters, according to sources familiar with the project. The model, internally called M3 Pro, is on track for release as early as the third quarter of this year.
This parameter count would surpass any other Chinese AI model currently known, marking a significant scale-up in the country's AI development efforts. The decision to open-source the model could accelerate adoption and research across the ecosystem.
At 2.7 trillion parameters, the M3 Pro would rival or exceed the size of some frontier Western models, though benchmarks remain unconfirmed. Open-source availability may also challenge the proprietary strategies of competitors like Baidu and Alibaba.
If MiniMax delivers on schedule, the model could provide a major boost to China's open-source AI community. However, export controls on advanced chips remain a potential bottleneck for training such large models at scale.
The Information's report, based on unnamed sources, has not been independently verified. MiniMax has not publicly commented on the project.