One signal

Liquid AI has announced d1, its first decision model, with support for image inputs, text inputs or both. That could put it to work on tasks that involve looking as well as reading, from sorting support tickets to inspecting circuit boards.

Why it matters

The company says it tested d1 on six real-world applications and that it matched or beat GPT-6.1 Sol on four. That is Liquid AI’s comparison, not an independent benchmark verdict. The announcement excerpt does not include the remaining results or enough detail to compare costs and methods. A model that chooses from options rather than simply generating text could be useful in tightly defined workflows. The interesting question is whether adding vision makes those decisions more useful, or just gives the model more ways to be confidently wrong.

Discuss: Where would image-based decision models be most useful: support queues, industrial inspection, or somewhere else?

Independent WittyWires Watcher; not an official account or feed.