Liquid AI releases d1-ultra for making decisions based on text and images
-mh23DVnFm2EDs4US6gJskniHIMwVY4.png)
Liquid AI has made d1-ultra available through Liquid API. It returns answer probabilities based on text and images. The price is 0.042 USD per million input tokens; output is not billed, but each question counts as a separate input, including all images.
Liquid AI has released the multimodal decision model d1-ultra, available through Liquid API. It accepts text, images or a combination of both, together with one or more questions. It returns answer probabilities without generating output tokens. It supports yes/no questions, multiple-choice questions and ratings on a scale.
The price is 0.042 USD per million input tokens; output tokens are not billed. Images are billed at the same rate as text and count as 1.5 tokens for each 32 × 32 pixel block. A 1024 × 1024 pixel image therefore represents 1 536 tokens. Each question is billed separately, including the associated text and all images.
Liquid AI cites checking printed circuit boards for defects and assessing a damage report based on a customer photo as examples of use cases. According to the company, d1-ultra achieves higher accuracy and better-calibrated probabilities than Jev 1.14 at the same price; the company describes a comparison with d1 and Jev 1.14 across 112 different evaluations. Demos are available in d1 Playground. Availability through Vercel and OpenRouter is still planned by the company.
Why it matters
Developers can integrate the model into tasks that require selecting an option or rating a situation based on text and a photo. To assess its suitability, it is important to check both the accuracy of its answers and the reliability of the probabilities it returns. At higher usage levels, costs will also be affected by the number of questions asked about the same material, because images are billed again for each question.
Release card
d1-ultra
Liquid AI
- Inputs
- The model accepts text, images or a combination of both and returns answer probabilities without generating output tokens. It supports yes/no questions, multiple-choice questions and ratings on a scale.
- Price
- $0.042 per million input tokens
- Availability
- The model is available through Liquid API, and demos can be tried in d1 Playground. Availability through Vercel and OpenRouter is still planned by the company.
- The model can be used to check printed circuit boards for defects against a specification.
- The model can be used to assess a damage report based on a customer photo.
- The model is unsuitable for generating free-form text responses because it returns probabilities without generating tokens.
Liquid AI says that d1-ultra achieves higher accuracy and better-calibrated probabilities than Jev 1.14 at the same price. Compared with d1, the company reports improvements particularly on knowledge tasks.
The card summarizes information from the article and any dated corrections, with a link to the original source. It is not our assessment of the model. It does not yet have a dedicated editorial profile. Model selection and other announcements →
Two audiences, two different impacts
What this means
For individuals
Through Liquid API, developers have access to a model for questions about text and images that provides probabilities for the available options instead of a written explanation. This determines which tasks it makes sense to try it on.
For a business
When designing automated photo checks, a company must budget for all images to be billed again for each question. Multiple checks on a single image therefore mean multiple billed inputs.
ProcessesCheck the original
Event sources
clearly official source · 1 publisher, 0 independent. We count feeds from the same owner only once.