MiniMax Drops M2.7: An LLM That Builds Its Own Training
Chinese AI startup MiniMax unveils M2.7, a proprietary model that constructs and optimizes its own reinforcement learning pipeline.
MiniMax just dropped M2.7, and the pitch is wild: it's a large language model that helped build itself. The Chinese AI startup calls it a "self-evolving" LLM, meaning the company used the model to construct, monitor, and fine-tune its own reinforcement learning harnesses.
That's a notable twist on standard AI development. Instead of relying entirely on human engineers to design training infrastructure, M2.7 played an active role in shaping the very systems that trained it. It's a feedback loop that could dramatically accelerate model improvement cycles.
MiniMax has steadily carved out a reputation as one of the more compelling players in the packed global AI landscape. M2.7 is proprietary, so don't expect open weights. But the self-evolving approach signals where frontier lab competition is heading — models that don't just learn, but architect their own learning.