MiniMax Open-Sources M3.1 Flash Model Weights with 428B Total Parameters
English text is being prepared.
Why read this
AI score 75
The original text provides details on the open-source weights, parameter scale, and inference efficiency optimizations, allowing readers to assess local deployment and integration costs.
Topics
Body · AI translation
MiniMax M3.1 Flash Open-Sources Model Weights, 428B Total Parameters
2026-10-23 09:08 IT Home - Wang Miao
IT Home, October 23: MiniMax (Xiyu Technology) announced today that it is open-sourcing the model weights of MiniMax M3.1 Flash, with 428B total parameters and 23B activated parameters.
Compared with the previous-generation M3 model, M3.1 Flash delivers an all-around improvement in model capabilities. In addition, through optimizations such as full-model sparse attention, quantization-aware training, and power-aware scheduling, M3.1 Flash further improves inference efficiency:
Full-model sparse attention: the first three full-attention layers are replaced with sparse attention;
QAT: routed MoE experts are migrated to W4A4 (excluding shared experts), and the sparse attention indexer and main branch are migrated from Q8KV8 to Q8KV4;
Power-aware scheduling: through fine-grained allocation of compute-card power consumption, compute cards are kept continuously at a high level of compute power, squeezing hardware efficiency to the utmost.
M3.1 Flash went live on October 19, and within 72 hours real users processed more than 100 trillion tokens.
MiniMax announced that it is joining hands with domestic and overseas chip manufacturers such as Huawei Ascend, T-Head, Kunlunxin, Moore Threads, Biren, Hygon, MetaX, Iluvatar CoreX, NVIDIA, and AMD, as well as inference and cloud service platforms such as Together AI, Fireworks, Nebius, Baseten, Modal, and DigitalOcean, to jointly advance day-one access for M3.1 Flash's open source.
In addition, inference frameworks and open-source communities such as vLLM, SGLang, BAAI FlagOS, Hugging Face, and ModelScope will also work with MiniMax to promote the model's application across different chips, cloud platforms, and development environments, continuously expanding the global AI open-source ecosystem.
IT Home attaches the open-source addresses below:
Advertising statement: External redirect links contained in this article (including but not limited to hyperlinks, QR codes, passcodes, and other forms) are used to convey more information and save selection time; the results are for reference only. All articles on IT Home that contain external links include this statement.