MiniMax M3.1 Flash weights released: 428B total, 1M context, license limits
M3.1 Flash weights are downloadable: 428B total/23B active, 1M context, vLLM/SGLang/TokenSpeed support; free non-commercial use, attribution for commercial, written approval above $20M revenue.
🇨🇳 MiniMax’s M3.1 Flash weights are actually out now. You can download them, not just try the model through an API. • Around 428B total parameters, with 23B active per token • Up to 1 million tokens of context • Takes text, images and video, and produces text • Serving support listed for vLLM, SGLang and TokenSpeed The license needs a closer look, though. Non-commercial use is free. Commercial users need to send a one-time notice and include attribution. If the relevant products or services generate over $20M in annual revenue, written approval is required first. And “23B active” doesn’t make this a laptop model. You still need serious hardware to run the full thing. A large Chinese model you can host yourself, with the weights available and the licensing strings attached. (https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash)