MiniMax Launches MiniMax M2.7, Initiating Model Self-Evolution
MiniMax has launched the MiniMax M2.7 model, achieving deep model participation in self-iteration for the first time. The model can construct complex Agent Harnesses and complete highly complex productivity tasks through capabilities such as Agent Teams and complex Skills. In the field of software engineering, M2.7 scored 56.22% in the SWE-Pro benchmark, approaching the level of Opus, and reached 55.6% in VIBE-Pro end-to-end project delivery. In the professional office domain, GDPval-AA achieved an ELO score of 1495, the highest among open-source models.
MiniMax has launched the MiniMax M2.7 model, achieving deep model participation in self-iteration for the first time. The model can construct complex Agent Harnesses and complete highly complex productivity tasks through capabilities such as Agent Teams and complex Skills. In the field of software engineering, M2.7 scored 56.22% in the SWE-Pro benchmark, approaching the level of Opus, and reached 55.6% in VIBE-Pro end-to-end project delivery. In the professional office domain, GDPval-AA achieved an ELO score of 1495, the highest among open-source models.
This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.