A specialist for pulling out information
Cloudflare’s October 19 announcement introduces Clairic, an early hosted preview for extracting values from text and images. This is useful news for AI geeks and experienced technicians building document or support workflows. The announcement day is visible; its exact publication hour is unconfirmed.
Developers specify fields such as an order number, date, or amount, rather than ask for a written reply. Text excerpts can include character offsets into the supplied text, giving an application a place to check the answer. Cloudflare describes confidence scores and a null option for missing values.
Hosted access, with limits
The official model page lists @cf/cloudflare/clairic on Workers AI at $0.10 per million input tokens. The changelog says there is no output-token charge. These rates do not establish the full cost of a working application. Cloudflare plans to release open weights later, without a date.
Cloudflare cautions that multi-step arithmetic and aggregation across many table rows remain difficult. Intervals can be too narrow on unfamiliar data, and forecasting is basic. Its speed, accuracy, and cost comparisons are provider reports; we have not tested the endpoint or reproduced those results.
Our take
R2’s editorial analysis: a valid date or number can still be the wrong one. Compare important fields with the original document, check missing-value handling, and let ordinary code calculate totals from checked inputs.
Start with a small set of representative examples whose answers you already know. Count wrong and missing fields, human corrections, and the cost of the complete workflow. A confidence score can help route review; it should not authorize a refund or payment on its own.