Skip to content

Pick your style: Denkbaar

Book a conversation

Articles

History

Qwen3.5 brings images and text together in an open model

·

Alibaba introduced Qwen3.5 with a multimodal design and a focus on execution costs.

An employee compares a card showing a mug with blocks on a laptop.
Check whether the image and description match. Illustration: Denkbaar.

On 16 February 2026, Alibaba introduced Qwen3.5. The announcement began with Qwen3.5-397B-A17B, a model able to process different types of information. The announcement names its multimodal design and Mixture-of-Experts architecture. In such a design, different parts of a model are used. At the introduction, Alibaba emphasised the cost of running applications. Source. An image card and a document lie beside a box containing separate blocks.

This mattered for the question of how a model becomes part of everyday software use. Images and text came together in the model’s design. For a business, the next step is to establish whether that combination helps with a concrete task and what is required to make the application work. Two colleagues compare a print of a mug with a description.

For a small or medium-sized business, a trial with your own image suitable for public use and a short description is appropriate. Ask the model which visible features match the description. Then check every observation and record what is missing or has been added. Ask a technical partner to explain the setup required for repeated use. Compare the output, human review and total costs with a simpler approach. Only then decide whether a multimodal model is the right building block for this task.

Sources

  1. Qwen3.5 open-sourced (397B MoE, native multimodal)

Topics: open sourcehistory