Can Models Also Be "Russian Dolls"? MiniMax Releases M2.7: The First Domestically-Produced Large Model Deeply Engaged in Self-Iteration

The speed of artificial intelligence evolution is shifting from “monthly updates” to “self-evolution.” On March 18, MiniMax officially released its first deeply iterative self-updating model—MiniMax M2.7. This marks a new stage in model development: large models are no longer solely fed by human programmers but are beginning to learn how to “bring themselves along.”

According to reports, the core breakthrough of MiniMax M2.7 lies in its highly autonomous construction ability. It can independently build complex Agent Harness (AI testing framework) and, relying on Agent Teams (collaborative AI), complex Skills, and tools like Tool Search, complete highly complex productivity tasks on its own.

In simple terms, M2.7 is not only a smarter conversational partner but also a “digital engineer” capable of self-diagnosis and self-optimization. This “self-participating iteration” mode will greatly enhance the model’s logical reasoning capabilities and tool invocation accuracy when facing unknown and complex tasks.

Currently, this self-evolving MiniMax M2.7 model is fully available on the MiniMax Agent platform and open platform. As large models begin to deeply participate in their own “growth,” the ceiling of AI may be pushed even higher.

Meanwhile, AI computing power and application markets are also evolving rapidly. LuCheng Technology announced the completion of a multi-billion yuan Series B funding round, with overseas revenue now soaring to 79%. Additionally, due to a surge in usage, some AI computing products from Alibaba Cloud have also announced price increases. Amid technological iteration and market volatility, the AI track in 2026 is becoming increasingly urgent and unpredictable.