Liquid d1.1: Decision API, Input Pricing and Local Alternatives

Send a defined question, receive a typed answer

The native API documentation uses model: d1.1 at https://api.liquid.ai/decisions/v1/systemone. Send a state and named questions using Choice, Score or Noul. The result includes an answers object and token usage; it generates no prose.

For example, define billing, technical and other as ticket categories. Test ambiguous examples as well as obvious ones, and route uncertain results to review. An answer probability is an input to your workflow, not proof that an action is safe to execute.

The $0.02 rate and repeated questions

Liquid's live price table lists d1 at $0.02 per one million input tokens, the same cached-input rate, and $0 output. It advertises $2 in starter credit per new organization. The price row says d1 while the current API docs say d1.1; verify the selected model's tariff in the console. This page does not infer a separate discount from the version name.

The usage rules charge each question for its text and all supplied images. Use the returned total input-token count when calculating cost.

Illustrative calculation: 1,000 inputs × 3 questions × 1,000 billed input tokens per question × $0.02 / 1,000,000 = $0.06. This assumes each question's complete billed prompt is exactly 1,000 tokens; use actual usage for a real bill. Zero output tokens does not make the input free.

Hosted and local versions have different contracts

The current family guide lists hosted d1.1 and d1-ultra, plus the separate open-weight d1-3B and d1-omni-600M. d1.1 accepts text and images through Liquid's API. Open weights for a related model do not mean the hosted d1.1 weights have been released.

For local deployment, select the exact open-weight model and review its runtime and license. The license terms summary distinguishes free use below the stated company-revenue threshold from enterprise licensing. Hosting, hardware and operations remain your costs.

Start with one question and check usage

Use a harmless labeled example, choose one output type, and inspect the answer and usage fields. Then add questions one at a time so you can see how the bill changes. If using images, follow the native API's accepted encoding and request limits rather than assuming another provider exposes the same route.

We reviewed official documentation and prices; no API call, latency test or accuracy benchmark was run. Compare deployment choices in the Decision Models guide and use the current request examples for implementation.