Chinese AI developer MiniMax intends to release a 2.7 trillion parameter model as open source later this year. The Information reports that two sources familiar with the plans confirm the project, which would surpass the size of any currently available Chinese model. Internally, the team refers to the project as M3 Pro, though the final name may change before launch. A release could occur as early as the third quarter, following the company’s current top model, M3, which contains 428 billion parameters. Larger architectures generally handle complex reasoning and multi-step instructions more effectively than smaller versions.
This move matters because it directly challenges the growing trend of Chinese open-source models gaining traction among developers seeking cost-effective solutions for high-volume tasks. MiniMax competes directly with Zhipu, DeepSeek, and Moonshot AI in a crowded market. However, the timing coincides with reports suggesting the Chinese government wants to tighten controls on future model releases. This potential regulatory shift could impact how widely these large models are distributed globally.
- The model is significantly larger than MiniMax’s current M3 offering.
- Regulatory pressure from Beijing may limit distribution channels.
- Competitors include Zhipu, DeepSeek, and Moonshot AI.
