Skip to content
System One
All models

Clairic

Provided by: system-one.dev

An extraction model that returns typed values from text, JSON and images.

Model info

  • Provided by: system-one.dev
  • 2 credits per 4KB of input
  • Inputs: text, JSON, and images (PNG, JPEG, WebP)
  • Images: up to 4 · 4MB each · 8MB total
  • Released 2026-10-19

Uses

  • Invoice field extraction: pull the amount, invoice date and invoice number from invoice text or a screenshot, usually with a confidence. See invoice and receipt extraction.
  • Order emails: pull the order number, quantity and delivery date; fields it cannot determine come back as null (when p_null exceeds null_threshold) rather than filled into value.
  • Ticket triage: pull the product name, error time and impact duration, then hand them to Jev or Clef to decide which team gets it.
  • Extract, then decide: Clairic gets the values, Jev or Clef decides among the options you list, and a generative model writes the reply if one is needed. See Clairic vs Jev and Clef.
  • Agent pre-processing: turn key values in a user message into structured data before the Agent routes or decides on them. For images, see screenshot to JSON.
  • Customer complaint email (an example from the official blog): from one complaint email, pull the order number, delivery date, total paid, days from delivery to the email, and a list of damage. The total paid is not written in the email; it is computed on the spot by adding 649 and 52.10 in tax to get 701.10. The 11 days are also worked out from two dates. Each damage item, "cracked screen" and "dented corner", comes with its location in the source text. The email has no tracking number, so that field returns null instead of an invented one.
  • Skewed, blurry receipt photo (an example from the official blog): from a receipt photo taken at an angle with hard-to-read text, pull the order number, purchase time, quantity, unit price, subtotal, tax and tip. The receipt shows no discount, so the discount field returns null. How this compares with OCR: OCR vs AI field extraction.
  • House price estimate from comparable sales (an example from the official blog): given a few comparable sales, an estimate field returns the subject property's median price and an 80% interval, p10 to p90. A narrow interval means the input supports the estimate well; a wide one means it is worth a second look.
  • Forecasting daily airport security-check passenger counts (an example from the official blog): pass the daily history as text and use a forecast field for the next day. It picks up the weekly pattern from the numbers alone, but not holidays or other events the series does not show. The official blog says forecasting is only for non-critical, rough pattern judgments and does not replace dedicated time-series models.
  • Three-step claim handling (an example from the official blog): Clairic extracts values from the email and uses an estimate field to give a range for the repair cost; Clef checks these against the policy and decides between refund and repair: repair is preferred when it costs less than 60% of the price and the damage does not stop the product from working, otherwise refund; and an LLM writes the reply. Step 2 sees structured values rather than the raw email, so it is less easily swayed by the wording of the email.

Official sources

Last updated 2026-10-21

Open GET /v1/models · Developer docs · Try live (sign in required)

Frequently asked questions

Is Clairic a large language model?
No. It does not generate text; it returns typed values.
What value types can Clairic return?
Nine: text, number, date, time, datetime, duration, list, estimate and forecast. Every answer has a value; confidence and p_null are optional (estimate and forecast fields, for example, don't include them). Estimates can optionally return 99 quantiles.
What does it return when a value is not there?
null, when p_null exceeds null_threshold; guess may still hold a value. Where p_null is present, it is the probability that the input can't determine this value.
What inputs does Clairic accept?
Text or JSON, plus up to 4 images (each up to 4 MiB, 8 MiB in total; remote image URLs are not supported). The context window is 65,536 tokens. There is no limit on the number of fields; fields that don't fit in one pass are split and processed in parallel. An optional reference_date sets which year a date written without a year belongs to. See the model info on this page for exact limits.
What are its limits?
The official blog says the preview is not yet reliable at long-chain calculations or multi-row aggregation, multi-step forecasts drift toward the average, and well-formed output is not necessarily correct. Test on your own samples before production, and set thresholds on confidence and p_null where they are returned.
How is Clairic different from Jev and Clef?
Clairic answers "what is the value"; Jev and Clef are decision models that answer "which one". Use them together: extract, then decide.
How is it billed?
How Clairic is billed: each 4KB of request content counts as one block, and anything under 4KB counts as one block; each block costs 2 credits. You're charged once per request, no matter how many fields you extract. This differs from Jev and Clef, whose cost is also multiplied by the number of questions. See pricing.
Can I try it online?
Yes: Try it online (sign-in required).