Liquid AI reports comparisons with d1 and Jev 1.14 across 112 evaluations and claims better-calibrated answers and higher accuracy than Jev 1.14 at the same price.
Thesis
d1-ultra makes a probability-bearing decision interface a product surface alongside generated-text models.
Observed facts
Liquid AI says d1-ultra accepts text, images, or both and returns probabilities without generating tokens.
1The model supports yes/no, choice, and score question types.
1Liquid AI lists API availability at $0.042 per million input tokens, with no output-token charge.
1Liquid AI says it compared d1-ultra with d1 and Jev 1.14 across 112 evaluations.
1
Mechanism
Rather than asking a language model to formulate a prose answer and then extracting a decision from it, the API takes one or more questions and returns probability values for bounded answer forms.
Why now
The release gives builders an immediately accessible API and a stated price, so this decision-oriented interface can be evaluated in concrete workflows rather than treated only as a research direction.
Implications
Teams with bounded classification or scoring tasks can test whether direct probability outputs reduce the need to generate and parse prose.
Evidence boundary
Those benchmark and calibration statements come from Liquid AI’s own announcement; the supplied source does not independently validate them or establish performance in production.
Watch conditions
Watch for independent evaluations of calibration and accuracy on representative real-world tasks, alongside confirmation of how API costs behave for multi-question and image-heavy inputs.