Cloudflare previews Clairic, a System One extraction model for typed values from unstructured inputs

Cloudflare announced a preview of Clairic, an extraction model hosted on Workers AI as @cf/cloudflare/clairic, with open weights planned “soon.” Clairic is designed to extract, compute, estimate, and forecast typed values from unstructured text and up to four optional images in a single pass, returning spans, confidence scores, p_null values, medians, and 80% intervals instead of generating tokens.

For builders using LLMs mainly to turn documents into structured fields, Clairic’s typed outputs, offsets, and no output-token billing could simplify validation and reduce cost for those specific workflows. The trade-off is scope: it does not write or reason like an LLM, and because this is a preview with open weights not yet released, teams still need to test accuracy, confidence thresholds, and fallback behavior on their own inputs.