Cloudflare released Clairic in Workers AI preview to extract names, dates and numbers from unstructured inputs. It predicts multiple fields together and returns confidence, null likelihood or estimated ranges, with billing based on input rather than output tokens. In the company’s example, a workflow using dedicated models for extraction and decisions was about 3.2 times faster than an entirely generative-model approach, but this is not a general performance guarantee. Multistep calculations and long-range predictions remain weaknesses, and correctly formatted values can still be wrong. Open weights are planned for the future.
Back to Tech
Cloudflare previews Clairic for extracting values from unstructured input
Separate extraction, decisions and text generation while retaining value checks and low-confidence review.