MiniMax-M3 vs Grok Build 0.1

MiniMax’s flagship, and the first open-weight model to combine frontier coding, a million-token context, and native multimodality in one place. M3 reads text, images, and video, and can operate a desktop computer. Multimodal training ran from step zero rather than being bolted on later. Weights are open. Built for long-horizon agent work that runs for hours rather than seconds.

xAI's coding model, purpose-built for autonomous software engineering rather than chat. Grok Build 0.1 acts as an engineering agent: it reasons through a problem, writes code, runs your terminal, checks for errors, and fixes its own mistakes in a loop. It reads text and images, always reasons before answering, and natively supports Model Context Protocol. Built for long-horizon coding that runs with little supervision, at a low cost, tuned for agent loops.