MiniMax releases open weights for 428B-parameter M3.1 Flash

MiniMax published the weights of M3.1 Flash on October 23, letting developers download and deploy the model. The mixture-of-experts model has about 428 billion total parameters with 23 billion activated per token, accepts text, image and video inputs, and supports a context window of up to 1 million tokens. It targets coding, AI agents and computer-use tasks.

• 428B total parameters, 23B activated per token
• Context window up to 1 million tokens with text, image and video input
• Faster inference via sparse attention and speculative decoding
• Deployment guides for SGLang and vLLM, chip support through FlagOS