Cloudflare launches Clairic, a System One extraction model

Cloudflare's introduction of Clairic comes as businesses deal with invoices, customer emails, and receipts that rarely share one format. Cloudflare describes the extraction work on such unstructured data as a place where AI is useful, but it points to drawbacks in general LLMs: they take time to generate an answer and can output numbers not actually present in the document. That creates a verification step and raises processing time and cost when data volume is large.

Clairic sits alongside Jev and Clef, which select an answer from prepared choices, while Clairic pulls concrete values, dates, and similar details from text and images. Cloudflare says Clairic is the first System One model to extend to extraction and numerical prediction, and it can also estimate or forecast based on input data — for example, inferring property prices from past real estate transactions or predicting next-day visitor counts.

In its airport security checkpoint test, Cloudflare reported an average error of 3.3% over 21 days, with actual counts falling within the prediction range on 17 of those days. The predicted range was exceeded on some days during the October 9–13 holiday stretch. Cloudflare says complex calculations and long-term forecasting remain challenges and that it plans to improve computing, estimation, and forecasting functions.