Skip to content
Rush Commerce
AI & Automation3 min read

Liquid AI d1.2 decision models add audio and data extraction

Liquid AI's d1.2 decision models add audio input and value extraction at $0.02 per million input tokens. Use the confidence score to route reviews.

Liquid AI released d1.2 and d1.2-ultra on November 16, the newest versions of its decision models. A decision model doesn't write text. It answers structured questions with a probability on every answer. The d1.2 release adds audio input and five new tasks, including reading values off invoices. The price is $0.02 per million input tokens for d1.2, with no output charge.

What actually happened

Per Liquid's announcement:

  • Two models. d1.2 at $0.02 per million input tokens, d1.2-ultra at $0.042. Both bill input tokens only.
  • How it works. You send the input (say, invoice text) plus a set of typed questions. You get structured answers back, each with a probability. A missing value comes back as "not found" instead of a guess.
  • New tasks. Read a value (text, number or date), find every span of a kind (like personal data), locate objects in an image, estimate a number with a range, and pick every option that applies.
  • Audio. Both models now take audio. Liquid reports 95% to 97% accuracy on sound-type and spoken-language questions in its own tests.
  • Benchmarks. Liquid says d1.2-ultra ranks first among decision models on the Artificial Analysis Decision Index v2.1, with a score of 75. The new-task accuracy numbers, like 96% to 98% on receipt values, come from Liquid's own test sets. Liquid notes that no public benchmark covers those tasks yet.
  • Availability. Both models are live on the Liquid API today. Liquid says OpenRouter and Vercel are coming soon. The post says nothing about open weights for d1.2.

Why decision models matter for your business

Most small-business AI work isn't chat. It's "what's the total on this receipt," "is this email a refund request," "does this photo show damage." Teams send those jobs to a general chat model, pay for output tokens, then parse free text with brittle code.

A decision model fits that work better. The answer is already structured. And the probability is the useful part: Liquid's pitch is that when the model says 80%, it should be right about 80% of the time. That gives you a dial. Auto-accept above your threshold, send the rest to a person.

Our caution: most of the accuracy numbers are Liquid's own. Before you move a workflow, run 200 of your real documents through it and compare against what your team actually keyed in. Calibration claims are only worth something if they hold on your data. We said the same about Cloudflare's Clef and Strands Decider: route the cheap, structured calls first, and measure.

Key takeaways

  • Liquid AI released d1.2 and d1.2-ultra on November 16
  • Pricing is $0.02 and $0.042 per million input tokens, with no output charge
  • New: audio input and five tasks, including value extraction and PII spans
  • d1.2-ultra ranks first among decision models on Artificial Analysis Decision Index v2.1; task accuracy figures are Liquid's own
  • Use the per-answer probability to set a human review threshold, and test on your own documents first

Most of your AI bill is structured work billed like chat. We build extraction and routing pipelines with confidence thresholds and a human review queue, then measure what you save. Run the numbers on your workflow, or send us a sample of your documents.

Sources: Liquid AI.

  • #liquid-ai
  • #decision-models
  • #document-extraction
  • #ai-automation
  • #llm-cost
TR

Tommy Rush — Founder, Rush Commerce

Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More

Get The Rush Report weekly — one email, zero fluff.