Cloudflare releases Clairic document data extraction model, faster and cheaper

Cloudflare released Clairic, a document data extraction model now in preview on Workers AI, with an open-weights version to follow. Users specify the fields they want (e.g., order number, invoice total), and the model returns the corresponding values in the correct format with confidence scores. Unlike language models, Clairic does not generate text token by token; it answers all fields at once, returning null for missing values. Beyond extraction, it supports simple calculations, estimation, and prediction. Cloudflare positions Clairic as one link in a larger agent chain alongside "System One" models like Jev and LLMs: for a complaint email, Clairic extracts the order number, amount, and damage details and estimates repair cost, decision model Clef chooses between refund and repair per company rules, and GLM-5.3 writes the customer reply. Cloudflare says this chain takes 19.3 seconds, versus 61.7 seconds and nearly double the cost for GLM-5.3 alone doing all three steps. For developers building document-processing pipelines on Workers AI, Clairic can serve as a low-cost front end for structured extraction; but Cloudflare acknowledges Clairic cannot reliably handle longer computation chains, and such logic must be executed in the developer's own code.