Clairic icon

Clairic

Clairic is Cloudflare’s System One extraction model for turning unstructured text and optional images into typed values, computed fields, estimates, and forecasts. It is hosted on Workers AI and is aimed at developers building structured data and AI workflows.

Clairic

Overview

Clairic is Cloudflare’s System One extraction model, hosted on Workers AI as @cf/cloudflare/clairic. It is designed to turn unstructured inputs into typed values rather than generate free-form text or choose from a fixed list of decisions.

A request combines source text, optionally up to four images, and named fields with declared types and optional instructions. Clairic can extract values directly, compute values from the input, estimate values within user-defined bounds, and forecast future steps in a series. Its single-pass design is intended for structured extraction tasks that need fast, typed outputs.

Key features

Text span extraction

Returns text spans copied from the input, including character offsets and a confidence score, for values such as names, order IDs, or descriptions.

Typed values

Reads or computes numbers, dates, times, datetimes, durations, and lists while honoring the declared type rules, such as decimal handling for currency amounts.

Bounded estimation

Produces bounded distributions for values not explicitly stated in the input. Users declare the minimum, maximum, and step grid, and the result includes a median and 80% interval.

Forecasting

Generates one estimate for each future step in a time series, with an 80% interval for each forecasted value.

Parallel multimodal requests

Accepts text plus up to four optional images and processes named fields in parallel. If the fields do not fit in one pass, the server can split them into parallel passes that each see the full input.

Confidence and null signaling

Returns typed values or null when it cannot find an answer, with p_null and confidence information for scalar outputs. It processes input tokens in a single forward pass and does not charge for output tokens, according to the article.

Use cases

  • Document and transaction extraction

    Read invoice or order documents and return amounts, names, order IDs, delivery dates, and other fields as typed values for downstream processing.

  • Computed operational fields

    Derive operational measures from source text, such as calculating how many days late a delivery was or computing a duration from stated dates and times.

  • Evidence-based estimation

    Estimate an unstated value, such as a repair cost, when the input provides evidence. The application can constrain the possible range and step grid and use the returned interval to assess uncertainty.

  • Short-horizon forecasting

    Forecast values beyond the observed input, such as traffic over the next 24 hours, with a separate uncertainty interval for each step ahead.

  • Structured AI pipelines

    Use extraction as an earlier stage in a broader workflow: convert raw material into typed values, pass those values to a decision model, and use an LLM later when free-form explanation or action language is needed.

Pros and Cons

Pros

  • Supports extraction, computation, estimation, and forecasting in one model family.
  • Produces typed and bounded outputs, including confidence information, null probabilities, and uncertainty intervals.
  • Processes requests in a single forward pass and does not generate output tokens, which the article positions as faster and less expensive for suitable tasks than traditional LLM calls.
  • Can combine text with up to four optional images in a field-based request.

Cons

  • The article describes Clairic as a preview, so production suitability and long-term availability should be evaluated before adoption.
  • The source does not provide hosted pricing, detailed setup instructions, accuracy benchmarks, or broader operational limits.
  • It is specialized for typed extraction, estimation, and forecasting; open-ended writing and deliberate reasoning remain outside its stated role.

FAQ

What is Clairic?

Clairic is a System One extraction model from Cloudflare. It reads text and can also process up to four optional images, returning typed values in a single parallel pass.

How does a Clairic request work?

A request contains input text, optionally up to four images, and a map of named fields. For each field, the user specifies a type and may add plain-language instructions. Clairic returns one typed answer per field; the source does not provide concrete invocation code or setup steps.

What kinds of outputs does Clairic produce?

It can return text spans, numbers, dates, times, datetimes, durations, lists, estimates, and forecasts. Scalar answers include a null probability and confidence, while estimates include a median and an 80% interval.

Where is Clairic available?

Clairic is hosted on Cloudflare Workers AI under the model identifier @cf/cloudflare/clairic. Cloudflare says it intends to release the model as open weights, but the source does not establish that this release is currently available.

When should Clairic be used instead of a general-purpose LLM?

Clairic is presented for tasks where an input needs to be converted into typed values, such as extracting invoice totals or delivery dates, computing lateness, estimating repair costs, or forecasting near-term traffic. The article does not provide a detailed accuracy benchmark, pricing schedule, or operational limits.

Quick Facts

Category
AI developer tool
Provider
Cloudflare
Hosted platform
Workers AI
Model identifier
@cf/cloudflare/clairic
Input
Text plus up to four optional images
Availability
Preview; Cloudflare says open-weights release is intended

Alternatives à Clairic