MiniMax-M2.7-Turbo

MiniMax-M2.7-Turbo is a speed-optimized serving of MiniMax M2.7, the 229-billion-parameter mixture-of-experts model MiniMax announced in March 2026 as the first of its models to take part in its own development. The Turbo label is a provider designation rather than a MiniMax release name, since MiniMax's own lineup offers M2.7 and M2.7-highspeed, and this build serves the model's native FP8 checkpoint with inference tuned for faster generation while keeping M2.7's reasoning quality. It inherits M2.7's strengths: 56.22% on SWE-Pro, 57.0% on Terminal Bench 2, 76.5 on SWE Multilingual, and 1495 Elo on GDPval-AA, along with native Agent Teams for multi-agent collaboration, a 97% adherence rate across more than 40 complex skills, and multi-turn editing of office documents. Function calling is supported within a 196,608-token context window. It suits latency-sensitive agent harnesses, production debugging, and high-volume coding loops where generation speed is the deciding factor. MiniMax recommends sampling at temperature 1.0, top_p 0.95, and top_k 40.