As of 08:59 Beijing time on October 23, 2026, open weights for MiniMax-M3.1-Flash are on Hugging Face at MiniMaxAI/MiniMax-M3.1-Flash (Hugging Face lastModified 2026-10-23T00:59:04Z). For API pricing, M Plan, MiniMax Code, and the Space Bunny note, see MiniMax M3.1 Flash.
What the model card says
The card describes a native multimodal mixture-of-experts model of about 428B total parameters with 23B activated per token, text, image and video inputs, text output, and a context window of up to 1 million tokens. The repo lists 48 model-*.safetensors shards. DigIntoAI has not re-run any of the card's benchmark tables.
License and local serving
The LICENSE file is the MiniMax Community License. Non-commercial use is free of charge under the stated conditions. For commercial use, the license requires prominently displaying "Built with MiniMax M3.1-Flash"; if products or services generate more than 20 million US dollars in yearly revenue, prior written authorization from MiniMax is required, otherwise a one-time notice to MiniMax is enough. The model card recommends serving the model locally with SGLang, vLLM or TokenSpeed.
MiniMax's release post on X also links ModelScope. Together AI hosts the model as a day-0 launch partner. MiniMax thanked Inferact for day-0 production work contributed back to vLLM, SGLang with NVIDIA and AMD for day-0 support, and NVIDIA NeMo for day-0 fine-tuning.
MiniMax's October 19 blog still says model weights will be released later, and that a Pro version will arrive later. The Hugging Face repo is live regardless; treat the blog wording on weights as stale until MiniMax updates that page. For MiniMax coding-plan context on DigIntoAI, see How to Use GLM, Qwen, MiniMax and Other Chinese Coding Plans Inside Claude Code.