Cloudflare has launched Clairic, a System One extraction model available on Workers AI. The tool is designed to extract, compute, and forecast typed values from unstructured data, such as invoices or repair costs. It supports various data types, including text, numbers, dates, and durations, providing confidence scores and probability metrics for each output.
The model processes requests containing text and up to four images. It handles unlimited fields by splitting inputs into parallel 8,192-token passes. For estimates and forecasts, Clairic provides a median value and an 80% confidence interval. In testing, the model successfully predicted TSA checkpoint counts with an average error of 3.3%.
Clairic is part of a three-part stack alongside decision models and LLMs. In a claims pipeline test, this architecture proved 3.2 times faster than using an LLM for all steps, reducing processing time from 61.7 seconds to 19.3 seconds. The cost per million claims dropped by 49%, from $11,740 to $6,032.
The model is currently available in preview via the Workers AI binding or REST API. Pricing is set at $0.10 per million input tokens with no additional output charges. While it is more expensive than the Jev model at $0.042, it is more cost-effective than the Clef decision model, which is priced at $0.24 per million tokens.
