HomeAI models
LiquidAI: d1.1
LiquidAI
•
d1.1
•
Released 
October 2026

LiquidAI
d1.1

Hosted decision model that returns calibrated probabilities for text and image inputs with zero generated tokens, for routing and moderation.

Modality:
Text
Image
model ID
liquid/d1.1
Start for free
arrow black icon
View API Docs
arrow white icon

Output Speed *

N/A
tok/s

Intelligence Index *

N/A
/ 100

Context Window *

65536
tokens

Input price

0.12
Anytoken

Output price

0
Anytoken
d1.1: Zero-Token Decisions for Routing, Moderation and Scoring d1.1 is Liquid AI's default hosted decision model and the successor to d1. It sits between d1 and the larger d1-ultra. It does not write text. It reads text and images in one forward pass and returns probabilities for yes/no, multiple-choice and scored questions, with output tokens always at zero. It suits high-volume routing, triage, guardrail checks and filtering, where calling a generative LLM for a one-word label is wasteful. Access the d1.1 API through AnyAPI.ai

Performance

Why d1.1 Targets the Gap Between d1 and d1-ultra

d1.1 is built for decisions where the answer is one of N known options. Liquid says it improves decision quality over d1 and answers with lower latency than d1-ultra, positioning it near d1-ultra's quality tier. Because it scores every option in a single forward pass with no decoding loop, latency does not grow with output length, and answers cannot fall outside your defined labels. For hot-path gates, such as routers and filters, this removes JSON parsing and schema retries. These are vendor claims, so validate accuracy on your own labeled traffic.

Technical Specifications

What the model supports

d1.1 accepts a state (text or JSON) plus optional Base64 images and one or more named, typed questions. It returns Noul, Choice or Score answers with probabilities. Two specifications matter most in production. First, output is a fixed answer type, not free text, so it replaces classifiers and LLM-as-judge calls but not generation. Second, it uses Liquid's System One request format, not OpenAI chat completions. Context window and knowledge cutoff were not stated in the sources reviewed.
Verified Specifications — 
d1.1
*
Input modalities
Text
Image
output modalities
Text
Context window
65536
 tokens
Maximum output tokens
58982
Reasoning
No
Knowledge cutoff
October 2026
Pricing (standard)
0.12
 AnyTokens in
 / 
0
 AnyTokens out

Limitations & Trade-offs

Where d1.1 falls short

1
No text generation. d1.1 returns only Noul, Choice or Score answers with probabilities. Liquid's documentation points to a language model for free-form writing, conversation, summarization, code generation and multi-step reasoning. It also cannot explain why it assigned a probability. If a downstream user needs a rationale or a new string, keep an LLM in the loop and use d1.1 only as the gate in front of it.
2
Limited public evidence. The sources reviewed give no d1.1 benchmark score, latency measurement or context window. Liquid positions it near d1-ultra's quality tier but advises calibrating decision thresholds on your own data. Matching schemas with d1 do not guarantee matching probability distributions, so rerun evaluation sets before moving production traffic or reusing thresholds tuned for another model.
3
Knowledge-dependent decisions. The original d1 showed weaker results on tasks that relied on external facts or deeper reasoning, and Liquid says d1-ultra's gains are largest on knowledge tasks. d1.1 may close part of that gap, but it is unproven. For fact-heavy moderation, complex policy checks or ambiguous multimodal cases, benchmark d1.1 against d1-ultra before committing.
4
Integration and billing constraints. Images must be Base64-encoded because remote URLs are rejected. Each question is billed as its own prompt, including its text and all images, so many questions about one large image multiply image-token usage. The request shape is not OpenAI chat-compatible, and the weights are not published, so teams that must self-host need an open-weight decision model instead.

Best-Fit Workloads

Where this model earns its place

01

High-volume routing and triage

‍

Liquid and gateway listings position d1.1 for routing, triage, intent detection and filtering on a hot path. One request can ask several typed questions, such as team, urgency and sentiment, about the same ticket, replacing sequential LLM classification calls. Output is always one of your defined options, so malformed JSON and schema retries disappear. Calibrate thresholds, for example escalating low-confidence tickets to a stronger model, on your own labeled traffic.

02

Moderation and guardrail gates

‍

Noul questions return a probability between 0 and 1, which makes policy checks threshold-driven. Liquid's moderation example blocks above one threshold, allows below another and sends the middle band to human review. d1.1 fits as a fast pre-filter on user input and agent actions. Ambiguous or knowledge-heavy policies should be tested against d1-ultra, since d1.1's accuracy on them is unpublished.

03

LLM-as-judge replacement and reranking

‍

Score questions rate an item on an ordered rubric and return a probability-weighted position, and Liquid lists reranking and LLM-as-judge replacement as best-fit use cases. For evaluation pipelines that score thousands of outputs, d1.1 avoids paying for generated verdicts and gives a continuous score instead of a self-reported number. Validate agreement with your human labels before replacing an existing judge.

04

Image-based inspection and screenshot decisions

‍

Liquid's docs show d1.1 answering questions about Base64 images, including comparing an inspected circuit board against a reference image in one request. This suits visual defect checks, image classification and decisions from screenshots without writing a text description first. Images are billed as input tokens at a fixed per-patch rate and re-billed per question, so ask only the questions you need. Use d1-ultra for harder visual cases.

Pricing in anytokens via AnyAPI
Input
0.12
₳
Output
0
₳
Cache write
—
₳
Cache read
0.12
₳
Get API access
arrow black icon
Read Docs

on this page

PerformanceTechnical SpecificationsLimitations & Trade-offsBest-Fit WorkloadsIntegrationFAQ

Integration

Access d1.1 via AnyAPI.ai

Access d1.1 through AnyAPI.ai using a unified API built for multi-model AI applications. Integrate d1.1 without maintaining a separate provider-specific connection, and keep the flexibility to test, switch, or combine models as your application requirements evolve.

01

One API integration

Access d1.1 and other AI models through the same API workflow instead of maintaining separate integrations for every provider.

02

Easy model switching

Test d1.1 against alternative models or switch models as your performance, capability, or cost requirements change without rebuilding your application around another provider API.

03

Flexible for production

Use d1.1 from experimentation through production while keeping your AI stack flexible as workloads, traffic, and model requirements evolve.

04

Multi-model applications

Use d1.1 for the workloads where it performs best and combine it with other models for tasks that require different capabilities, performance, or efficiency.

Start free
arrow black icon
Read Docs

Frequently Asked Questions

Answers to common questions about integrating and using this AI model via AnyAPI.ai

d1.1 is Liquid AI's default hosted decision model and the successor to d1. It takes a state (text or JSON, optionally images) and typed questions, reads them in one forward pass, and returns probabilities for yes/no, choice or score answers. It generates no text. It is positioned between d1 and d1-ultra for routing, moderation, reranking and LLM-as-judge replacement.

No. d1.1 returns typed decision answers with probabilities and reports zero output tokens on every request. Billing covers input tokens only. Because there is no decoding loop, latency does not grow with output length, and answers always match the question type, so malformed JSON is not a failure mode.

Yes. Liquid's documentation lists vision as available with d1.1 and d1-ultra. Each image must be a Base64 data URL or a content_type/base64 object, and remote image URLs are not accepted. Images are counted as input tokens at 1.5 tokens per 32×32-pixel patch, and each question is billed with all supplied images.

Use d1.1 for high-volume decisions where cost and latency matter most. Use d1-ultra for harder, knowledge-dependent or high-stakes decisions. Liquid describes d1-ultra as its most capable decision model and says d1.1 has lower latency than d1-ultra while approaching its quality. Test both on your own labeled data.

A d1.1 context window was not stated in the Liquid sources reviewed, so none is given here. Image tokens count toward each question's input token limit, and every question is billed with its text and all images. Test representative state lengths and image counts before production, and check Liquid's current API documentation for limits.

* Benchmark data source: Artificial Analysis artificialanalysis.ai