New models · Liquid AI · decide API

Liquid AI d1 on Jev AI

d1 is Liquid AI’s family of decision models. Give it a state and typed questions and it returns a probability for yes/no, one option with a probability for each, or a level on your rubric, read from the model instead of written out as text. On Jev AI, d1.1 and d1-ultra take the same request as Jev, so a playground run, saved judge or API call switches to them by changing the model field.

Try Liquid d1.1 Create API key Copy API request

Checking connection…

Try it on this page

Run Liquid d1.1 right here

Edit the example or paste your own text, then run it. Sign in to run: new accounts get 20 welcome credits, and a failed run costs nothing.

Open the full playground

Run with

Support ticket triage · Liquid d1.1

1 Input

Hi, this is the third time I am writing. Our payouts have failed every day since Monday and my team cannot pay suppliers. I already re-entered the bank details twice. If this is not fixed today we will have to move to another provider.

2 Questions

Choice Which team should handle this ticket?
Yes / No Does the customer need a response today?
Score How frustrated is the customer?
Yes / No Does the customer threaten to leave or switch providers?

2Hosted d1 models on Jev AI: d1.1 and d1-ultra

65,536Tokens of context on d1.1 (32,768 on d1-ultra)

#5d1-ultra on the Benchmark Heaven API board, of 48 ranked

What d1 is

Liquid AI builds d1 on its LFM models and serves the hosted versions through OpenRouter’s Decisions API. A d1 model does not chat or write text: it evaluates every question against the shared state and returns typed answers. Its question types are Jev’s own: noul for yes/no, choice for one of your options and score for a level on an ordered rubric.

On Jev AI it runs alongside Jev, Clef, Laya, Mercury Decide, GPT-6 Luna and RightNow System Zero. Jev AI sends every d1 request to Liquid AI through OpenRouter and never answers it with another model. If d1 is unavailable, the request fails before any charge. This first release reads text and JSON; images are not offered on Jev AI yet.

d1.1 or d1-ultra

Both models answer the same questions with the same request. Pick d1.1 for long inputs and d1-ultra when you want Liquid’s largest hosted decision model.

Model	d1.1	d1-ultra
Model field on Jev AI	d1.1	d1-ultra
Decisions API alias	liquid/d1.1	liquid/d1-ultra
Context per request	65,536 tokens	32,768 tokens
Released by Liquid AI	October 27, 2026	October 15, 2026
Benchmark Heaven API board	Not yet measured	#5 of 48
Jev AI model rate	1×	1×

d1.1 succeeds the first hosted d1, which OpenRouter retires on November 20, 2026. Jev AI never offered the first d1 and does not map its name to d1.1, so an old request fails instead of silently switching models.

How d1 compares with Jev

Benchmark Heaven’s API board (JevBench v1.7.48) ranks hosted decision APIs on one composite of intelligence, calibration and speed. d1-ultra is #5 of 48, the first hosted d1 is #13, and Jev 1.13 is #15.

Model	API rank	Composite	Intelligence	Calibration	Speed
Liquid d1-ultra	#5	76.0	69.7	91.3	88.9
Liquid d1 (first release)	#13	72.8	62.1	86.8	87.9
Jev 1.13	#15	71.4	63.6	90.6	91.2
Jev 1.14	#18	70.2	59.9	90.1	91.0

d1.1 is not on the board yet, so these rows do not tell you how it compares. Run your own examples on d1.1, d1-ultra and Jev side by side in the playground: the request is the same.

On d1, the state is the costly part of the input, and the questions add only their own text. In our test on November 2, 2026, the same state reported 2,141 input tokens with one question and 2,179 with two.

Limits and billing

d1.1 and d1-ultra use 1× the Jev model rate: 1 balance token per input token on the API, the same as Jev. Output tokens are free.
You pay for the input tokens the request reports: the state plus each question’s instructions and options. With 2,179 input tokens, a two-question request costs 2,179 balance tokens.
Playground and credit-fallback runs cost 1 credit, as with Jev.
Input over the model’s context is rejected without a charge. If Liquid AI or OpenRouter is unavailable, the request returns an error before any charge.

View API pricing API reference All models

Use d1 with your Jev API key

Create a key on the Jev API page and store it as JEV_AI_API_KEY on your server. Use the Jev endpoint with d1.1 or d1-ultra, or the Decisions-compatible endpoint with liquid/d1.1 or liquid/d1-ultra.

curl --fail-with-body https://jev-ai.pro/api/v1/systemone \
  -H "Authorization: Bearer $JEV_AI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "d1.1",
  "state": "Customer: I was charged twice for order 4821 and want my money back today.",
  "questions": {
    "team": {
      "type": "choice",
      "instructions": "Which team should handle this message?",
      "criteria": {
        "billing": "Payments, charges and refunds",
        "shipping": "Delivery problems",
        "technical": null
      }
    },
    "urgent": {
      "type": "noul",
      "instructions": "Does this need a reply within the hour?"
    }
  }
}'

Read answers.<id>.noul for the probability of yes, .choice for the selected option and .score for the rubric level, exactly as with Jev.

Open weights: d1-2B and d1-Omni-2B

Liquid AI also publishes two small d1 models on Hugging Face under the LFM Open License v1.0. Jev AI does not host them; this is a guide for running them yourself.

Model	Parameters	Built on	Context	Reads
d1-2B	1.92B	LFM3-VL-2B	128K	Text, JSON, pictures and video
d1-Omni-2B	2.17B	LFM3-Omni-2B	128K	Adds audio and spoken questions

Liquid says answers get less reliable past about 64K tokens of state, although the context is 128K.
The license allows free commercial use for companies under US$10M in annual revenue; read the full license before you ship.
Liquid AI reports 77.9 for d1-2B on JevBench v1.4.1. That is Liquid’s own test on an older JevBench version, not a Benchmark Heaven result, and Jev AI has not reproduced it.
Neither model is ranked on Benchmark Heaven’s boards yet.

Run d1-2B with Transformers (Liquid’s own model code):

pip install "transformers>=5.19" torch torchvision pillow

import torch
from transformers import AutoModel

model = AutoModel.from_pretrained("LiquidAI/d1-2B", trust_remote_code=True, dtype=torch.bfloat16).to("cuda")
out = model.system_one("I was charged twice for order 4821.", {
    "refund": {"type": "noul", "instructions": "Is the customer asking for a refund?"},
})
print(out["answers"]["refund"]["noul"])

Questions

Do I need a Liquid AI or OpenRouter account?

No. Jev AI connects to Liquid AI through OpenRouter for you, so one Jev API key and one balance cover d1.1, d1-ultra and every other model on the site.

Can I send images to d1 on Jev AI?

Not yet. This first release reads text and JSON. For images, use Jev 1.14, Clef, GPT-6 Luna or RightNow System Zero.

Does every question pay for the state again?

Not on the usage OpenRouter reports today: the state is counted once per request and each question adds its own tokens. Jev AI bills the reported input tokens, so if that changes, your bill follows the reported usage.

Is d1.1 better than Jev?

Benchmark Heaven has not measured d1.1 yet. d1-ultra ranks above Jev 1.13 on the API board, at #5 against #15. The fastest answer for your workload is to run the same examples on both in the playground.

Can I run d1-2B on Jev AI?

No. d1-2B and d1-Omni-2B are open weights you run yourself. On Jev AI, use the hosted d1.1 or d1-ultra.

Sources

Liquid AI, LFM and d1 are trademarks of Liquid AI. Jev AI is not affiliated with or endorsed by Liquid AI. Details about Liquid’s models are from their public model cards and OpenRouter pages as of November 2, 2026 and can change.