MiniMax M2
Follow the version trail of the MiniMax M2 open-weight line, and see how a lean MoE activation budget serves coding workflows and what each step from M2.1 to M2.7 changes for tool calling and office editing.
MiniMax M2 is the open-weight model series MiniMax builds for agentic and coding work, covering M2, M2.1, M2.7 and M2-HER-2. It uses a Mixture-of-Experts design of roughly 230B total parameters with about 10B active per token, cutting inference cost while strengthening tool calling and multi-file edits. Later releases add multilingual programming and Office editing.
