Liquid AI Gives d1 Image Input, Matching GPT-6.1 Sol At Lower Cost
A model that answers in a single forward pass without generating tokens sidesteps the cost structure of generative systems, so matching a frontier model on four of six tasks at a fraction of the price is the opening bid in a different kind of comparison than benchmark leaderboards use.
Reporting from 1 source: GIGAZINE.
Liquid AI's decision model d1 now accepts images as well as text or both at once, the company said. Tested against GPT-6.1 Sol and Claude Opus 5.5 on six applications, from filtering support tickets to inspecting circuit boards, d1 matched or beat GPT-6.1 Sol on four, ran at 19 to 200 times lower cost, and answered faster on every task, at 200 to 300 ms for text.
d1 does not write out answers. It returns the probability of each option in one forward pass, which is why text decisions land in 200 to 300 milliseconds and why image input is a cheap addition: the same pass now reads pixels or prose. Liquid AI says the model was tested on six real applications, from support ticket filtering to circuit board inspection, against GPT-6.1 Sol and Claude Opus 5.5.
On demos, d1 classified good and defective parts at 85 to 97 percent accuracy without training, read a Wordle board straight from a screenshot so developers no longer needed a text version, and guessed drawings in Quick, Draw! at an average of 5.2 out of 6.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.