Cloudflare previews Clairic, a cheap model that pulls values from documents

Cloudflare has launched a preview of Clairic, a model built to pull structured values out of messy text and images, such as an invoice total, a delivery date or a list of items, and to estimate or forecast numbers from the evidence in the input. You give it the text plus a set of named, typed fields, and it returns one typed answer per field in a single pass, each with a confidence score and, where a value is missing, a null. Because it does not generate text token by token, Cloudflare says it is faster and cheaper than using a general chatbot for the same job, and cannot produce malformed output. It is hosted on Workers AI at $0.10 per million input tokens with no charge for output, within a free allowance of 10,000 neurons a day, and Cloudflare says open weights will follow. Cloudflare notes this is an early checkpoint and warns that multi-step arithmetic and forecasting are weak, so lean on the confidence scores to decide what to automate.