Alibaba Launches Qwen3.8-Flash With Major Efficiency Gains
Alibaba’s Qwen team has released a new multimodal AI model focused on coding, office tasks and lower-cost inference, while also previewing technology for the upcoming Qwen4 generation.
Alibaba has launched Qwen3.8-Flash, its latest AI model designed to deliver strong performance while significantly reducing the cost of training and running advanced AI systems. The company says the model improves particularly in coding and general office tasks, strengthening Qwen’s position in an increasingly competitive global AI market.
Qwen3.8-Flash uses a Mixture-of-Experts (MoE) architecture with 125 billion parameters, while only around 6 billion parameters are activated for each token. Alibaba says this architecture allows the model to achieve stronger capabilities with substantially lower computational requirements. Compared with Qwen3.7-Plus, the company reports that training Qwen3.8-Flash-Next required roughly one-ninth the training cost while producing better results in coding and office-related tasks.
Another major feature is its ability to process extremely large amounts of information. Qwen3.8-Flash supports a 262,144-token context window by default, with the ability to extend this to 1 million tokens. This could make the model useful for analyzing large documents, extensive codebases, lengthy conversations and other information-heavy workloads.
Alibaba has also released the weights for Qwen3.8-Flash-Next, an experimental version that provides an early look at the architecture planned for the company’s future Qwen4 family. The model combines several architectural improvements, including Gated DeltaNet, Qwen Sparse Attention, Gated Residuals and N-gram embeddings. Alibaba says the release allows developers and researchers to evaluate the architecture before the full Qwen4 family arrives.
The launch comes shortly after Alibaba committed billions of dollars toward expanding its AI infrastructure. The company recently announced an approximately $10.2 billion Hong Kong share placement, saying the proceeds would support its AI strategy, including computing infrastructure, chips and model development. Qwen3.8-Flash therefore represents not only a new model release but another step in Alibaba’s broader effort to compete with leading AI companies worldwide.
