MiniMax: MiniMax M3

MiniMax M3 is a multimodal model from MiniMax that accepts text, image, and video inputs, making it one of the fewer models that can process video alongside other modalities. It supports tool use and reasoning, and offers a context window of 1,048,576 tokens, which is large enough to handle lengthy documents or extended conversation histories without truncation. Structured output support is unconfirmed from available data. At $0.30 per million input tokens and $1.20 per million output tokens, MiniMax M3 sits at the budget end of the market. Its blended benchmark score of 78.2 covers only 3 benchmarks, so that figure should be treated as preliminary rather than definitive. The coding subscore of 96.6 is notable and may appeal to developers, while the agentic score of 58.3 suggests more caution for autonomous workflow use cases. Teams that need video input at a low cost have few alternatives and may find this worth evaluating.