MiniMax to open-source 2.7 trillion-parameter model M3 Pro

MiniMax plans to open-source a new large language model with 2.7 trillion parameters later this year and could release it as early as Q3, the Information told two people familiar with the plans. Internally the company calls the project "M3 Pro," though the name could change before launch.

What is MiniMax building and when will it arrive?

MiniMax is developing a 2.7 trillion-parameter language model, internally called M3 Pro, and aims to publish it as open source, with a release that could come as early as Q3. The project would be larger than any Chinese AI model currently on the market and would follow the company’s current top model, M3, which has 428 billion parameters.

Work on much larger parameter counts reflects a common engineering bet: larger models tend to perform better on tasks that require complex reasoning and multi-step instructions. The Information is the outlet cited for the plan, which it attributed to two people familiar with the matter.

How does M3 Pro compare to MiniMax’s current model and rivals?

M3 Pro at 2.7 trillion parameters would be more than six times larger than MiniMax’s existing M3, which the company lists at 428 billion parameters. Chinese open-source models have gained traction with developers this year, especially among those seeking cheaper models for high-volume, less critical tasks, and MiniMax competes with Zhipu, DeepSeek, and Moonshot AI.

Making a much larger model available as open source would change how developers can access high-capacity Chinese language models, since most similarly sized systems are either commercial or closely held. The plan to publish the model as open source, if executed, would therefore alter availability and competition within that developer ecosystem.

Why it matters

A 2.7 trillion-parameter open-source model would raise the floor for what developers in China can run without vendor lock-in, and it would intensify competition with existing players named in the report: Zhipu, DeepSeek, and Moonshot AI. At the same time, the company’s timeline and decision to open-source a very large model intersect with a political backdrop: recent reports suggest the Chinese government wants to tighten controls on future releases of such models, which could affect distribution or deployment.

What to watch

Watch for a public release or repository announcement in Q3, and for any regulatory guidance from Chinese authorities addressing large-model releases. Confirmation of an open-source license, an official model name, and the timing of a public code or weights drop will determine whether this plan changes developer access in practice.