The Qwen3.8-Flash release and its pricing

Alibaba's Qwen team released Qwen3.8-Flash on August 26, 2026, a multimodal Mixture-of-Experts model built around cost per token, alongside open-source weights for a sibling release called Qwen3.8-Flash-Next. Qwen said it would charge 1 yuan (about $0.1488) per million input tokens and 3 yuan per million output tokens for access through its application programming interface. The model supports a default context window of 262,144 tokens, expandable to 1 million tokens, which Qwen said enables handling of large files, lengthy conversations and extensive research materials. Compared with its previous Qwen3.7-Plus model, Qwen said Qwen3.8-Flash requires about one-ninth of the training cost.
Architecture details and Qwen4 positioning
Qwen3.8-Flash-Next pairs a 125B main model with 51B N-gram embedding parameters and a 4B multi-token prediction module, totaling 180B parameters on disk, with only 6B parameters activating per token. Four changes carry the release: a hybrid attention layer combining Gated DeltaNet and Qwen Sparse Attention, a Gated Residual branch structure, an N-gram embedding table, and the Muon optimizer applied alongside AdamW to specific weight categories. The FP8 checkpoint is 172.78 GiB and the BF16 checkpoint is 335.28 GiB, with TP2 as the minimum validated FP8 configuration on GB300 and TP4 recommended. Qwen positioned the release as an early preview of the architecture that will underpin Qwen4, the same role Qwen3-Next played for Qwen3.5.
Strategic context
Alibaba framed Qwen3.8-Flash as delivering stronger coding and office-task performance while cutting training costs, releasing open-source weights for Qwen3.8-Flash-Next so developers can evaluate an architecture Qwen said would serve as a prototype for its next-generation Qwen4 model family. Separately, Alibaba shares rose 2 percent on August 26, 2026 after the model introduction, according to one report on the day's market reaction. The release comes as AI has become the largest driver of revenue growth for Alibaba, where competition in China's domestic AI sector is intensifying.
Share this article







