Independent assessments show that MiniMax M2.7 rivals leading closed-source models in core intelligent agent tasks.

ME News message, April 5 (UTC+8). Recently, according to LangChain’s independent evaluation, the open-weight model MiniMax M2.7 has reached a level comparable to that of closed-source leading models on core agent tasks such as file operations, tool calling, and instruction following. The article’s view is that its advantage is that costs have been reduced by about 20x, and speed improved by 2–4x. Taking an example of producing 10 million tokens per day: using Opus 4.6 costs about $250/day, while MiniMax M2.7 only needs about $12/day. This progress is driven by open benchmarks such as SWE-Rebench and Terminal Bench 2.0, making open models a feasible option for deploying agents in production environments and enabling them to be used in combination with closed-source models. (Source: InFoQ)