[ Compare ]
Jev vs Liquid AI d1-ultra: same price, higher score?
Published By Jev Hub
[ Sponsors ]Launch price
These products help keep Jev Hub updated. Want yours up here?
- Liquid AI's d1-ultra, announced on 15 October 2026, answers the same three kinds of question as Jev with probabilities and no text, and lists the same price: $0.042 per million input tokens, with no charge for output.
- It also takes images.
- Liquid says it beats Jev 1.14 on all of its own 112 evaluations, and Artificial Analysis's independent Decision Index scores it 65.2 against 57.1 for Jev 1.14 and 57.5 for Jev 1.13, the highest of the 13 models listed.
- That index is days old and tests four things, not your task.
- Choose Jev for the longer-available service and its published track record; look hard at d1-ultra if image input or the higher index score matters.
Side by side
| Feature | Jev (TypeSafe AI) | d1-ultra (Liquid AI) |
|---|---|---|
| What it is | A decision model: it answers typed questions about an input and returns probabilities. It cannot write text. | A multimodal decision model: it takes text, images or both, plus one or more questions, and returns probabilities in one pass without generating tokens (Liquid AI's announcement). |
| Question types | Choice (up to 255 options), Score (2 to 10 levels), Noul (yes/no) | Yes/no (noul), single-label choice and scale scores |
| Input | Text by default; images are an experimental option (since 14 October) | Text and images |
| Input price | $0.042 per million tokens, output free (checked 14 October 2026) | $0.042 per million input tokens; output is not billed |
| How questions are billed | Questions about the same input go in one request, so the input is read once (TypeSafe's quick start) | Each question is billed as a separate prompt that includes its text and images (Liquid AI's announcement) |
| Images in the price | Image tokens count as input; TypeSafe labels images not ready for production | A 1024 by 1024 image counts as 1,536 input tokens (1.5 tokens per 32 by 32 patch) |
| Where you run it | TypeSafe's API, Vercel AI Gateway, OpenRouter and LiteLLM | Liquid AI's API now; Liquid says it will be available on Vercel and OpenRouter soon |
| Weights | Closed | Built on a "larger open-weight model"; Liquid's page does not say whether d1-ultra's own weights are released, and press coverage reports API-only |
| Vendor benchmark claim | TypeSafe's own workflow evals | Liquid says it beats Jev 1.14 on all tasks across 112 evaluations, with better calibration, at the same price (Liquid's own test) |
| Independent score (Artificial Analysis Decision Index, 18 October) | 57.5 (Jev 1.13) and 57.1 (Jev 1.14) | 65.2 (d1-ultra), the highest of the 13 models listed; its smaller d1 scored 50.3 |
The d1-ultra column uses Liquid AI's announcement of 15 October 2026 and Artificial Analysis's leaderboard read on 18 October. The Jev column uses TypeSafe's documentation. Prices and limits can change.
The short answer
d1-ultra is the first decision model to beat Jev on an independent leaderboard. It answers the same kinds of question, lists the same price and takes images as a normal feature. On Artificial Analysis's Decision Index it scored 65.2, against 57.1 for Jev 1.14 and 57.5 for Jev 1.13.
That is a real gap, but one index, days old, with four tests that are not your task. It does not tell you which model is better at yours, and it is a measure of accuracy-style scores, not of speed, cost in your workflow or how well it holds up in production.
What does Liquid AI say?
Liquid AI announced d1-ultra on 15 October 2026 and calls it the best decision model. On its own page it says d1-ultra outperforms Jev 1.14 on all tasks across 112 evaluations, with better calibration and higher accuracy at the same price, and that it beats every other decision model on the Artificial Analysis Decision Index by at least 8 points. The evaluations are Liquid's own and the page gives no scores in text, so the first claim is a vendor claim.
The second claim matches what Artificial Analysis shows: 65.2 for d1-ultra, and 57.5 for the next best, Jev 1.13.
What is different in how you use it?
The three question types are the same. Two documented differences matter in practice. First, images: d1-ultra takes them as a normal feature, while Jev's image input is experimental and TypeSafe labels it not ready for production. Second, billing for several questions: Liquid says each question is billed as a separate prompt that includes its text and images, while with Jev the questions about one input go in one request and the input is read once. For several questions on a long input, that can change the real cost even at the same list price.
What is not known yet?
Before you rely on any of this:
- Weights. Liquid says it is built on a larger open-weight model, but its page does not say whether d1-ultra's own weights are released.
- Context size and languages. Liquid's announcement page does not state them.
- Speed. The page gives no measured latency. Artificial Analysis also reports time and cost, but we have not confirmed those figures and do not quote them.
- Availability. It is on Liquid AI's API now, with Vercel and OpenRouter promised soon.
Which should you pick?
Based on what is documented today:
- Choose Jev if you want the service with the longest track record and the most independent tests, or you batch many questions on long inputs.
- Look at d1-ultra if you need image input as a normal feature, or the higher Decision Index score is the thing you care about.
- Test both before you commit. Run the same labelled sample of your own data through each, and compare accuracy, speed and total cost, including how each bills several questions.
Keep reading
More on how Jev compares:
Common questions
A multimodal decision model announced on 15 October 2026. It takes text and images, answers yes/no, choice and score questions, and returns probabilities without generating text.
On Artificial Analysis's independent Decision Index it scored higher: 65.2 against 57.1 for Jev 1.14 and 57.5 for Jev 1.13. That is one new index with four tests, so check it on your own data before switching.
The list price is the same, $0.042 per million input tokens with no output charge. Liquid bills each question as a separate prompt, so a request with several questions can cost more than the same questions batched on Jev.
Yes. Liquid says a 1024 by 1024 image counts as 1,536 input tokens. Jev's image input is experimental.
Liquid's announcement says it is built on a larger open-weight model but does not say whether d1-ultra's own weights are released. Press coverage describes it as API-only.
Sources
Written by Jev Hub from the sources below and checked against them on . Where a figure is a company's own claim, the article says so.
Related content
- Read the article
Compare
Published
Jev vs Snowflake Decision
Snowflake's decision model runs inside SQL. It is not open to everyone yet.
- Read the article
Compare
Updated
Jev vs ChatGPT, Claude and other LLMs
ChatGPT, Claude and Gemini side by side: what each does, what it costs, which to use.
- Read the article
Compare
Updated
Jev vs ChatGPT: which one do you need?
A chat model writes; Jev decides. What it costs, side by side.
