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

MiniMax released the open weights for its M3.1 Flash model on October 23, following the model's initial service launch on October 19. The mixture-of-experts model has approximately 428 billion total parameters with 23 billion activated per token, supports text, image and video inputs, and offers a context window of up to 1 million tokens. MiniMax says M3.1 Flash, aimed at coding, AI agents and computer-use tasks, improves inference efficiency through sparse attention, quantization-aware training and speculative decoding, and the release includes deployment guidance for SGLang and vLLM plus multi-chip support through FlagOS.

MiniMax released the weights for its M3.1 Flash model on October 23, making the model available for developers to download and deploy. The mixture-of-experts model has approximately 428 billion total parameters, with 23 billion activated per token. It supports text, image and video inputs and a context window of up to 1 million tokens. Designed for coding, AI agents and computer-use tasks, M3.1 Flash improves inference efficiency through sparse attention, quantization-aware training and speculative decoding. The release includes deployment guidance for inference frameworks such as SGLang and vLLM, alongside support for multiple chip platforms through FlagOS. The weights follow the model’s initial service launch on October 19. MiniMax https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash