The pace of artificial intelligence's evolution is transitioning from "monthly updates" to "self-evolution." On March 18, MiniMax officially launched its first version of the model that deeply participates in its own iterative process—MiniMax M2.7. This marks a new stage in model development: large models are no longer merely fed by human programmers, but are beginning to learn to "raise themselves."

According to the introduction, the core breakthrough of MiniMax M2.7 lies in its strong self-building capability. It can independently build complex Agent Harness (agent testing framework) and complete highly complex productivity tasks by relying on underlying capabilities such as Agent Teams (agent collaboration), complex Skills, and Tool Search tool.

In short, M2.7 is not only a smarter conversationalist, but also a "digital engineer" capable of self-diagnosis and self-optimization. This "self-participation iteration" mode will greatly enhance the model's logical reasoning limit and tool calling accuracy when facing unknown complex tasks.

Currently, this self-evolving MiniMax M2.7 model has been fully launched on the MiniMax Agent platform and open platforms. When large models begin to deeply participate in their own "growth" process, the ceiling of AI may be pushed higher again.

At the same time, the AI computing power and application market have also seen frequent dynamics. LuChen Technology announced a round B financing of several hundred million yuan, with its overseas revenue accounting for as high as 79%; and due to a surge in usage, some AI computing power products of AliCloud have also reported price increases. Under the interweaving of technological iteration and market fluctuations, the AI race in 2026 is becoming increasingly urgent and full of uncertainty.