Frontier-level performance at 10x less pricing than Opus and OpenAI. That’s MiniMax M2.7 - shipped just 5 weeks after M2.5.

But here’s the real story: M2.7 is proprietary. Closed weights.
The model literally helped build itself. It ran 100+ autonomous loops of debugging, code fixes, and eval analysis, handling 30-50% of its own RL research workflow.
MiniMax joins Zhipu (GLM-5 Turbo) in ditching open source. Alibaba’s Qwen team just lost its technical lead and there are rumors Qwen3.5-Max will be locked behind paid APIs.
Chinese AI labs spent the last year being the open-source heroes. Now the best ones are closing the gates.
The open-source golden age from China might already be ending.