MiniMax reportedly develops 2.7T parameter M3Pro

Open-sourcing a 2.7T parameter model is a bold, high-stakes play to dominate the agentic AI developer ecosystem, but the real test will be whether the inference costs make it practical for actual deployment.

• –A 2.7-trillion-parameter model will require unprecedented computational resources for inference, making quantization and community optimization efforts critical.
• –A heavy focus on reasoning and multi-step agents suggests MiniMax is building this model specifically to handle complex agentic workflows rather than general-purpose chat.
• –If successful, this release could force other major AI labs to accelerate their own open-source plans to remain competitive in the developer space.