Cloudflare released a preview of Clairic on 19 October, PPC Land reported. Available on Workers AI as @cf/cloudflare/clairic, the model reads text and images and returns typed values, including numbers, dates and lists, rather than writing sentences.
A request names the fields to fill and supplies the material to examine. Cloudflare says Clairic processes that input in a single forward pass and generates no output tokens. Each scalar answer carries a confidence score and a p_null value indicating the probability that the requested value is absent. In Cloudflare’s customer-email example, it calculated an amount from a price and tax, counted the days since delivery and returned null for a tracking number that the email did not contain.
Checking a claim against an email could therefore involve asking for several values at once, while leaving a missing detail empty rather than filling it with a guess. In Cloudflare’s test, the three-step customer-claim pipeline was 49% cheaper and about 3.2 times as fast with Clairic as it was when run entirely on a large language model. Cloudflare describes the preview endpoint as an early checkpoint that it expects to improve.
.jpg){kind=link}