AI
Alibaba's Qwen team releases open-weight 125B multimodal model trained at one-ninth the cost
The Qwen team at Alibaba released Qwen3.8-Flash-Next on 27 August, a 125-billion-parameter mixture-of-experts model that activates roughly 6B parameters per token and is described as a preview of the Qwen4 architecture. The model handles text, images and video natively, ships with a 1-million-token context window, lists pricing of $0.15 per million input tokens and $0.50 per million output tokens, and is released under MIT with weights on Hugging Face.