Is d1-ultra a drop-in replacement for Jev?

For the request format, yes. Through OpenRouter’s System One endpoint or JevStation it accepts the same state and the same Choice, Score and Noul questions and returns answers in the same shape. d1-ultra also accepts images, but JevStation sends text only for now.

Is d1-ultra really the best decision model?

The #1 ranking (a score of 65 on the Artificial Analysis Decision Index) and the 112-evaluation suite are Liquid AI’s own figures; we found no independent reproduction yet. On our 196 zero-shot inputs it was level with Jev, Microsoft-Decision-1 and d1, and on 2026-10-07 Clef-flash scored higher on Banking77 (97.9%). Where it did stand out was calibration.

What is the difference between d1-ultra and d1?

d1 is Liquid’s first hosted decision model (2026-09-30), at $0.02 per million input tokens and a 65,536-token context. d1-ultra is the larger model, at $0.042 and 32,768 tokens. In our run they had the same accuracy and counted exactly the same tokens; d1-ultra was more careful with its confidence (0 errors in 41 answers at 0.99 or above, against 2 in 104 for d1).

Why does d1-ultra cost 2 credits on JevStation?

Because it reads the state once per question, like Perplexity Decider. Twelve long questions on 24,000 characters came to 52,975 input tokens, against 8,981 for Jev. At 1 credit that request would cost us more than our margin floor allows, so d1-ultra is 2 credits standard and 6 large. d1, at half the list price, stays at 1 and 3.

Can I try d1-ultra without an account?

Yes. d1-ultra and d1 are both in the free no-signup trial on this page.

Jev vs Liquid AI d1-ultra: is the new #1 decision model worth switching to?

Liquid AI released d1-ultra on 2026-10-15 and says it ranks first on the Artificial Analysis Decision Index. It takes the same request as Jev, a state plus typed questions, at the same list price of $0.042 per million input tokens, and it is already on OpenRouter. In our run on 2026-10-16 it was level with Jev on accuracy (87.5% against 86.5% on Banking77 card intents, 91% against 89% on AG News) and was the best calibrated model we have tested. The catch is cost: d1-ultra reads the whole state again for every question, so a five-question request on a long text uses about four times the tokens Jev does.

Updated 2026-10-16

Try it on your own text

Edit the example and run it. No account needed — 3 free runs a day.

Scenario

Route a ticket to the right team, rate its urgency and flag churn risk.

Text to judge166 / 2,000

Edit freely — the model answers the scenario's questions about whatever you put here.

Model

What Jev actually returned

This is the unedited output of the example above, captured from the live API.

Hi, I've been trying to connect my Stripe account for 3 days and it keeps failing with 'invalid redirect'. I'm losing sales every hour this is down. Please help ASAP.

urgency

score

How urgent is this ticket?

3.73/4Critical

4 · Critical73%
3 · Urgent27%
0 · Can wait0%
1 · Normal0%
2 · Soon0%

Confidence 77%

churn risk

noul

Is the customer at risk of churning?

Yes80%

Yes 80%No 20%

Captured 2026-10-09 · 0.89 s · jev-1.13.0

At a glance

(vendor) is the company selling the model; (ours) is a call we made through OpenRouter on 2026-10-16.

	Jev	d1-ultra	d1
Made by	TypeSafe	Liquid AI	Liquid AI
Model id	`jev-latest`	`liquid/d1-ultra`	`liquid/d1`
Released	—	2026-10-15	2026-09-30
List price, input	$0.042 / M tokens	$0.042 / M tokens	$0.02 / M tokens
Output tokens	Free	Free	Free
Context	~32k tokens, shared with the longest question (vendor); 262k opt-in	32,768 tokens (vendor)	65,536 tokens (vendor)
Images	Opt-in preview (vendor), not on JevStation	Yes (vendor), text only on JevStation	Yes (vendor), text only
Languages	Not stated	44 (vendor)	Not stated
Input tokens, our fit	≈ 260 + 0.163 × characters + 43 × questions	≈ questions × (37 + 0.169 × characters)	Same as d1-ultra
Median latency	378 ms (ours)	525 ms (ours)	533–548 ms (ours)
Banking77 card intents	86.5% (ours)	87.5% (ours)	86.5% (ours)
AG News topics	89% (ours)	91% (ours)	91% (ours)
Credits on JevStation	1 / 3	2 / 6	1 / 3

Credits are standard / large; an evaluation is large above 8,000 characters of state or five questions.

Accuracy and calibration

We sent all three models the same labelled inputs on the same day, zero-shot, with one line of description per option. The method is on Decision models compared, which runs the other models on the same data.

Task	Jev	d1-ultra	d1
Banking77, 12 card intents (96)	86.5%	87.5%	86.5%
AG News, 4 topics (100)	89%	91%	91%
Banking77: answers at ≥ 0.9 confidence	78% of inputs, 96.0% right	65% of inputs, 96.8% right	73% of inputs, 100% right
Wrong at ≥ 0.99 confidence	3 of 144	0 of 41	2 of 104
Expected calibration error, Banking77	0.065	0.034	0.065

Accuracy is a tie: one or two points on a hundred items is noise. Calibration is where d1-ultra is different. It rarely says 0.99 (41 of 196 answers, against 144 for Jev) and was never wrong when it did, and its Banking77 calibration error is about half that of the others. If you auto-act above a high threshold, that is the property that matters. The trade is coverage: fewer answers clear the bar, so more go to a human.

All three found a single cancel request hidden anywhere in 20,000 characters of neutral text (d1-ultra: 0.991 at the very end, 0.001 on the control).

Tokens and cost

The list price of d1-ultra equals Jev’s, so the bill follows the token count, and here the two models behave very differently. Jev reads the state once; d1-ultra and d1 read it once per question. We fitted both on nine requests (state of 100, 2,000 and 8,000 characters, with one, three and five questions):

Request	Jev tokens	d1-ultra tokens	Cheaper at list price
1 question, short ticket (~300 chars)	~350	~90	d1-ultra, about 4×
3 questions, 1,500 characters	~640	~870	Jev, about 1.4×
5 questions, 8,000 characters	~1,780	6,964	Jev, about 3.9×
12 long questions, 24,000 characters	8,981	52,975	Jev, about 5.9×

d1 counts the same tokens at half the price, so it is cheaper than Jev up to about three questions on a 1,500-character text. For one question per call, both Liquid models are among the cheapest options we have measured: $0.005 per thousand AG News calls for d1-ultra and $0.002 for d1.

Latency was a little slower than Jev, with a p50 near 530 ms and p90 under 700 ms on both tasks. We saw one 3.5 s response in about forty requests and no rate limits.

When to pick which

Auto-acting above a strict confidence threshold: d1-ultra, the best calibrated model in our runs; confirm on your own labels.
One question per call, at volume: d1, the lowest cost per call we have measured, at the same accuracy.
Many questions about a long text: Jev, or Microsoft-Decision-1 or Luna Decisions, which read the state once.
Image inputs: d1-ultra supports them through Liquid’s API or OpenRouter; JevStation does not pass images yet.
Highest accuracy on fine-grained intents: Clef-flash led Banking77 in our 2026-10-07 run (97.9%).

Try them

Run d1-ultra or d1 above without an account, or sign in and use the side-by-side runner to send the same text and questions to Jev and d1-ultra at once. The pricing calculator shows the cost of each for your own numbers.

Sources

d1-ultra launch, multimodal input, 44 languages and Decision Index claim. Liquid AI blog and launch post (vendor)
d1 and d1-ultra price and context. OpenRouter: d1-ultra and d1 (vendor)
No independent overall score yet. BenchLM (independent)
Jev price and context. TypeSafe model docs (vendor)
Accuracy, calibration, latency and token counts. Our runs through OpenRouter on 2026-10-16 (ours)

JevStation is independent. Jev, d1 and d1-ultra are names of their respective owners, TypeSafe and Liquid AI; we are not affiliated with either.

Frequently asked questions

Is d1-ultra a drop-in replacement for Jev?: For the request format, yes. Through OpenRouter’s System One endpoint or JevStation it accepts the same state and the same Choice, Score and Noul questions and returns answers in the same shape. d1-ultra also accepts images, but JevStation sends text only for now.
Is d1-ultra really the best decision model?: The #1 ranking (a score of 65 on the Artificial Analysis Decision Index) and the 112-evaluation suite are Liquid AI’s own figures; we found no independent reproduction yet. On our 196 zero-shot inputs it was level with Jev, Microsoft-Decision-1 and d1, and on 2026-10-07 Clef-flash scored higher on Banking77 (97.9%). Where it did stand out was calibration.
What is the difference between d1-ultra and d1?: d1 is Liquid’s first hosted decision model (2026-09-30), at $0.02 per million input tokens and a 65,536-token context. d1-ultra is the larger model, at $0.042 and 32,768 tokens. In our run they had the same accuracy and counted exactly the same tokens; d1-ultra was more careful with its confidence (0 errors in 41 answers at 0.99 or above, against 2 in 104 for d1).
Why does d1-ultra cost 2 credits on JevStation?: Because it reads the state once per question, like Perplexity Decider. Twelve long questions on 24,000 characters came to 52,975 input tokens, against 8,981 for Jev. At 1 credit that request would cost us more than our margin floor allows, so d1-ultra is 2 credits standard and 6 large. d1, at half the list price, stays at 1 and 3.
Can I try d1-ultra without an account?: Yes. d1-ultra and d1 are both in the free no-signup trial on this page.

Make it part of your pipeline

Create free account Read the API docs