Cloudflare Launches Clairic, Faster than LLMs According to the Company

Cloudflare unveils Clairic, an AI model dedicated to field extraction from unstructured documents. The company highlights time and cost savings compared to using a standalone LLM in a typical scenario, while acknowledging limitations with longer calculations. Clairic is available in beta on Workers AI, with open weights announced as forthcoming.

Cloudflare reports 19.3 seconds for the chain, 61.7 seconds for GLM-5.3 alone

Cloudflare states that the combination of Clairic, the Clef decision model, and GLM-5.3 processes a claims email in 19.3 seconds. In contrast, GLM-5.3 executing the three steps alone takes 61.7 seconds and, according to the company, costs nearly twice as much. Cloudflare clarifies that Clairic does not yet reliably handle longer calculation chains and currently recommends executing these calculations in the application code.

Clairic available in beta on Workers AI, open weights expected

Cloudflare has unveiled Clairic, an AI model it describes as capable of extracting information from documents more quickly and at a lower cost than large language models. Clairic can be used in preview on Workers AI. The company also plans to release open weights soon.

Field extraction, one pass per field, and integration into agent chains

Users specify the fields to extract, such as an order number or an invoice total, and Clairic returns the associated values in the appropriate format, along with a confidence score. Unlike language models, Clairic does not produce text incrementally token by token: it provides a response for each field in a single operation and returns null when a value is missing. In addition to extraction from unstructured text, it is capable of performing basic calculations, estimates, and forecasts. Clairic is designed to integrate into an agent chain, alongside LLMs and System One-type models like Jev. In the given example, Clairic extracts the fields and estimates the repair cost, Clef applies the rules to choose between reimbursement and repair, and then GLM-5.3 drafts the response to the customer.