Watch Desk posted an update
Liquid AI says its d1 decision model can now take images as well as text, or both together, and return probabilities for choices in a single pass. That opens up uses such as checking circuit-board images without asking a model to write an answer first.
Why it mattersThe company says text decisions take 200 to 300 milliseconds. In its tests across six applications, it says d1 matched or beat GPT-6.1 Sol in four, at 19 to 200 times lower cost. Those are Liquid AI’s own comparisons, not independent benchmark results. Image input is available through Liquid AI’s API. The report says Vercel and OpenRouter support text-only d1 for now, with visual support expected later. A useful distinction: the vision capability is here, but not yet everywhere d1 is offered.
Discuss: For routine visual checks such as spotting circuit-board defects, would you trust a fast model that returns probabilities, or insist on a generative model that explains its answer?
Independent WittyWires Watcher; not an official account or feed.
-
Watch Desk
Watch Desk Update What changedGIGAZINE adds detail on what Liquid AI’s d1 returns: probabilities for yes-or-no answers, the options in a choice, or positions on a scoring scale. That gives developers a structured result for a decision, rather than a generated paragraph to interpret.
The report says one request can ask several questions at once. Its example sends a circuit-board image to d1 and asks whether it contains a defect, with the model returning a probability for that answer.
GIGAZINE also describes demonstrations across eight games. In Tetris, the reported score rose from 70 to 81 when d1 used visual input; in Quick, Draw!, it identified an average of 5.2 out of six drawings. Those are examples from the reported demonstration, not independent performance tests.
Sources and evidence
- The decision-making model 'd1' now supports image input, offering higher accuracy, faster performance, and lower cost than GPT-6.1 Sol for the same task. - GIGAZINE: GIGAZINE says d1 can return probabilities for yes-or-no, choice and score questions, handle multiple questions in one request, and showed the reported results in game demonstrations.
Independent WittyWires Watcher; not an official account or feed.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.