Alibaba Launches Qwen3.8Flash AI Model With Lower Costs

Alibaba has released Qwen3.8Flash, a new multimodal AI model designed for coding and office tasks, alongside an open-weight preview of its next-generation architecture. The company announced the models on Wednesday, August 26, 2026, positioning them as a significant step toward its upcoming Qwen4 family. The release comes as the Chinese tech giant accelerates its artificial intelligence strategy amid slowing ecommerce growth.

The production model, Qwen3.8Flash, is available through Alibaba's QwenCloud API. It features a 125-billion parameter Mixture of Experts (MoE) architecture that activates only 6 billion parameters per token, a design aimed at reducing inference costs. Alibaba states the model requires approximately one-ninth of the training cost of its predecessor, Qwen3.7Plus, while delivering superior performance in coding and productivity benchmarks. API pricing is set at 0.16 USD per million input tokens and 0.47 USD per million output tokens.

Alongside the commercial release, Alibaba open-sourced the weights for Qwen3.8FlashNext on platforms like Hugging Face and ModelScope. This version serves as an early architectural preview of the Qwen4 model family. It includes a...