Model comparison · pricing checked Sep 30, 2026

The same decision workflow, with a different billing path

Jev and Liquid d1 both evaluate shared state against typed questions. Their request and response semantics are compatible, while their gateway formats and current prices differ.

This app currently calls Jev through OpenRouter and Liquid d1 through Vercel AI Gateway. The shared playground can run the same state and questions against either model after a small server-side adapter maps Noul to boolean and normalizes usage.

Compare the operating details

Compatible decisions, provider-specific wire formats

Use the same examples to compare behavior. This page keeps the price snapshot separate from model-quality claims, which still need a labeled evaluation set.

Input price

Jev

$0.042 / 1M tokens

Liquid d1

$0 / 1M tokens

Jev uses the OpenRouter rate configured by this app. Vercel currently lists Liquid d1 input and output at $0.

Provider price pages

50% gross-margin credit rate

Jev

840 / 1M input tokens

Liquid d1

0 / 1M tokens + 1 / request minimum

At $0.0001 per credit, a 50% margin charges twice upstream cost. The zero-price d1 rate therefore computes to zero before the per-request minimum.

Decision output

Jev

Typed answers with probabilities

Liquid d1

Typed answers with probabilities

Both return boolean, choice, and score decisions. The adapter translates field names and token-usage casing; the Gateway response does not add a separate confidence field for Choice or Score.

Vercel evaluation API

Pick a model with a shared test set

Start with price and integration, then measure answer quality on your own cases.

01

You want to try d1 through the Vercel AI Gateway.

Use Liquid d1

The site routes d1 to Vercel’s evaluation endpoint using the configured VERCEL_AI_GATEWAY_API_KEY.

02

Your existing app already calls Jev through OpenRouter.

Keep Jev in the comparison

The current Jev integration remains available; choose either model in the shared workbench without changing the question editor.

03

A confidence threshold will trigger an action.

Evaluate before routing

Run labeled examples through both models and compare false positives, false negatives, and calibration before you set production thresholds.

What changes when the provider changes?

The playground stays the same; server-side routing adapts each provider contract.

Jev · OpenRouter Decisions API

Noul, Choice, and Score

The current integration sends the shared state and Jev’s typed questions to OpenRouter, then bills input tokens at the configured app rate.

  • Upstream input price is $0.042/M and output is $0/M.
  • The current app rate is 600 credits/M, equal to $0.06/M or about 30% gross margin on model-token cost.

Liquid d1 · Vercel AI Gateway

Boolean, Choice, and Score

The gateway evaluation API accepts the same shared state and typed-decision workflow. This app maps its Noul question to boolean and converts the returned probability into the Jev playground answer shape.

  • Vercel currently lists $0/M for both input and output, with a 32K context and 0 generated output tokens.
  • The 50% gross-margin rate is 0 credits/M; successful calls still meet the existing one-credit minimum.

Request and response compatibility

The decision contract matches after a small provider adapter.

State and questions
JevShared text or structured state; Noul, Choice, Score
Liquid d1Shared text or structured state; Boolean, Choice, Score
Yes/no answer
Jevanswers[id].noul
Liquid d1answers[id].probability; mapped to Jev’s noul field
Choice and Score
JevProbabilities and confidence
Liquid d1Probabilities; no separate confidence field
Token usage
Jevusage.input_tokens / output_tokens
Liquid d1usage.inputTokens / outputTokens; normalized by the server
Current token rate
Jev$0.042/M input; 600 credits/M billed here
Liquid d1$0/M input and output; 1-credit request minimum

Sources and calculation

Provider prices were checked on September 30, 2026. The 50% gross-margin target uses sell price = model cost ÷ (1 − 0.50), with $0.0001 per credit. Jev’s 840 credits/M is the target rate; the current app rate remains 600 credits/M. Liquid d1 is currently listed at $0/M, so its target token rate is 0 and the app minimum of 1 credit/request determines the actual charge. Because listed token cost is zero, that minimum yields more than a 50% margin on model-token cost alone; Gateway, payment, and serving costs are excluded.

Try the same case in the playground

Liquid d1 ausprobieren

Führe d1 im gemeinsamen Entscheidungs-Workbench aus und wechsle dann zu Jev, um denselben Zustand und dieselben Fragen zu vergleichen.

Was möchtest du entscheiden?

Wähle eine Aufgabe, um Eingabe und Regeln zu laden. Beispiele zu laden kostet kein Guthaben.

Liquid d1 ↗Span · Verhaltensbewertung ↗

Text oder JSON, das für jede Frage separat ausgewertet wird.

Erweiterte Fragen-Einstellungen

Fragen 1/8

Which team should handle this ticket? Use other when no option fits.Choice

Optionen

Maximales Guthabenbudget: 1 credits

Wir reservieren dieses Budget, bevor das Modell ausgeführt wird, rechnen anschließend den tatsächlichen Verbrauch ab und erstatten die Differenz. Fehlgeschlagene Aufrufe werden erstattet.

Neues Konto? Erhalte einmalig 100 Guthabenpunkte. Kurze Anfragen verbrauchen meist 1 Guthabenpunkt; längere Eingaben können mehr kosten.

10 Startguthabenpunkte · ein Guthaben für Web und API

Entwurf in diesem Browser gespeichert · Dein Entwurf wird nach der Anmeldung wiederhergestellt.

Die Konfiguration bleibt während der Einrichtung in diesem Browser-Tab.

BEISPIEL · KEIN GUTHABEN VERBRAUCHT

„Mir wurde heute Morgen zweimal Geld abgebucht. Bitte erstatte die doppelte Zahlung noch heute.“

Eine Eingabe → strukturierte Entscheidungen

Choice

Abrechnung

An das Abrechnungsteam weiterleiten

Ja / Nein

96%

Wahrscheinlichkeit einer dringenden Anfrage

Score

2.8 / 3

Dringlichkeit auf einer Skala von 0 bis 3

Statische Darstellung des Ausgabeformats. Die Werte sind weder eine Live-Modellantwort noch ein Maß für die Genauigkeit.

Dein Entwurf wird nach der Anmeldung wiederhergestellt.