Cloudflare has announced a preview of Clairic, a System One-style model it describes as extracting, computing, estimating and forecasting values from unstructured inputs. The company says Clairic is hosted on Workers AI as @cf/cloudflare/clairic and can be called from a Worker or through a REST API. It plans to release the model’s weights later.
A developer supplies text, up to four optional images and named fields with types such as text, number, date, list, estimate or forecast. Cloudflare says Clairic returns an answer for each field in one pass. Copied text includes character offsets; estimates include a median and an 80% interval. Scalar answers also carry a confidence score and p_null, the model’s estimated probability that the value is absent from the input.
In Cloudflare’s customer email example, $649 plus $52.10 in tax produces a $701.10 total, while a tracking number missing from the email returns null. The example illustrates the proposed output, not independently tested accuracy.
Values rather than choices
Cloudflare distinguishes Clairic from its description of Jev and Clef, decision models that return probabilities for listed options. Clairic instead returns extracted, computed or predicted values. Cloudflare says its single-pass approach generates no output tokens and carries no output-token charge.
In one company-reported comparison, a customer-claim pipeline using Clairic, Clef and GLM-5.3 took 19.3 seconds and cost $6,032 per million claims. An all-GLM-5.3 version took 61.7 seconds and cost $11,740 per million claims. Those figures describe Cloudflare’s example pipeline, not a general performance guarantee.
