Qwen in one minute
Qwen can support text processing, programming and image or video analysis. Specialised models handle tasks such as image generation or speech. On 8 October 2026, the range includes the Qwen3.8 generation, with commercial Max and Flash models and the open-weight Qwen3.8-27B model.
Max, Flash and open models
Alibaba’s model catalogue lists Qwen3.8-Max and Qwen3.8-Flash. Max is a product line for demanding tasks; Flash emphasises a practical balance of speed, cost and capability. These are names within a family, not a universal quality ranking.
Qwen3.8-27B also provides downloadable weights under Apache 2.0. Developers can run it on their own infrastructure. This differs from a managed API, where the provider handles execution under its service terms. Check each model’s licence: one open model does not make the entire commercial family open.
A model name in a cloud catalogue may be a moving alias. For a business application, record the version or snapshot too, so changes in answers can be investigated.
Text, images, audio, video and tools
Qwen includes models for writing, reasoning and programming. Current vision-language models can combine text with images and video inputs. Understanding a video and producing a new one are different tasks. Alibaba’s separate Wan range should not automatically be presented as a capability of every Qwen model.
Qwen-Image models handle image generation; speech offerings include Qwen-Audio services and Omni variants. Available inputs and outputs depend on the model and access method. Start with the task: describe a photograph, create an image, transcribe audio or produce a spoken reply.
For AI Agents, a model is combined with tools and an execution environment. Qwen Code is a development tool for programming tasks. The language model can propose an action, while the application determines which files or services it can actually access.
Example: a multilingual product catalogue
A wholesaler wants Dutch product descriptions based on supplier documents and photographs. A suitable multimodal Qwen model can structure product data, match images to products and flag unclear information. A subsequent step prepares shop copy without inventing missing certifications or dimensions.
A staff member checks the drafts against their sources. Only then does a separate CMS integration publish the approved data. Measure how many fields are correct and how many still need correction. A managed service can be practical for a small catalogue; self-hosting requires technical capacity and suitable hardware. Open weights is a decision about access and management, not an automatic saving.