Liquid AI has taught d1, its decision model, to look at pictures. By the company’s own count, the upgraded model drew level with or passed GPT-6.1 Sol in four of six applications, which means it fell short in two.
A decision model does not write replies. You hand it options, such as yes or no, or a list of categories, and it returns a probability for each in a single pass, with no text generated. That design explains the low price and quick response. It also means d1 cannot draft, explain, or chat the way a conversational model does.
Liquid AI reports costs 19 to 200 times below GPT-6.1 Sol and Claude Opus 5.5, with faster answers throughout. Liquid AI quotes 200 to 300 milliseconds per answer on text inputs. The company chose the six applications, which include sorting support tickets and inspecting circuit boards. No outside benchmark backs any figure.
Teams with high-volume classification work should rerun their own samples before switching, since a four-of-six record on someone else’s tasks says little about yours.
Liquid AI (@liquidai) on X, 5 October 2026.
