A photo becomes a decision signal
Liquid AI’s d1-ultra is presented as a multimodal decision model. In the company’s circuit-board example, the model compares an image with a question about component defects and returns probabilities: 0.80 for a missing component, 0.11 for an extra one and 0.10 for a scratch.
That output is more actionable than a free-form description: a quality-control pipeline can route a high-scoring case for inspection. But a probability is not proof. Production use would still need calibrated confidence, representative images and a human or downstream check for consequential decisions.
What the price and evaluation do—and do not—tell you
Liquid says the model was evaluated on 112 tasks and lists input pricing of $0.042 per million input tokens. Those are company-published details. Builders should confirm current availability and pricing, then test their own images: lighting, camera angle, board revisions and rare defect types can change results.
Source published 2026-10-15. Coverage is based on the maker’s announcement and demonstration.