MiniMax M2.7

MiniMax M2.7 is a text model from MiniMax aimed at autonomous, real-world productivity: complex software engineering, agentic tool use, and office document workflows. Its headline feature is "self-evolution" — MiniMax says an internal version of the model autonomously optimized a programming scaffold over 100+ rounds, analyzing failure trajectories, modifying code, running evaluations, and deciding whether to keep or revert changes, for a reported 30% improvement on internal benchmarks. It also introduces native Agent Teams for multi-agent collaboration with stable role identity and autonomous decision-making.

Within the same family, M2.7 follows MiniMax M2.5 and precedes MiniMax M3 and its M3 Preview. MiniMax reports significant gains over the previous generation in professional finance tasks — for instance, autonomously reading annual reports and earnings calls, designing assumptions, and building revenue models like a junior analyst.

On vendor-reported evaluations, M2.7 scores 56.22% on SWE-Pro, 76.5 on SWE Multilingual, 55.6% on VIBE-Pro, and 57.0% on Terminal Bench 2, alongside a reported 1495 ELO on GDPval-AA. These figures are self-reported by MiniMax rather than independent evaluators.

The model supports reasoning, code-optimized generation, function calling, and web search, with a context window near 198K tokens. MiniMax describes system-level uses such as correlating monitoring metrics, trace analysis, and SRE-style debugging, citing live incident recovery reduced to under three minutes on multiple occasions. It is distributed under a non-commercial MiniMax license.

Sources

MiniMax M2.7 - Model Self-Improvement, Driving Productivity Innovation Through Technological Breakthroughs | MiniMax· minimax.iominimax-m2.7 Model by Minimaxai· build.nvidia.comMiniMaxAI/MiniMax-M2.7 · Hugging Face· huggingface.co

This About section is AI-generated from public sources via VeniceStats + Venice inference, with no human editing. It may contain inaccuracies.