MiniMax M3

MiniMax M3 is a multimodal foundation model that processes text, image, and video inputs to produce text output, with a 524K token context window.

Strengths

Handles long contexts and supports multiple input modalities including video, making it capable of processing diverse content types in a single request.

Best for

Extended-context tasks like video analysis, long-document coding work, and agentic systems that need to reason over mixed media.

Limitations

Limited to text output only, so it cannot generate images or video; performance on specialized domains may vary compared to dedicated single-modality models.