Cloudflare's Clairic Reads Invoices Without Generating Text
Extraction at low cost and a null for missing fields is the part an LLM pipeline struggles with, and Cloudflare is pitching that gap rather than another general model.
Reporting from 1 source: GIGAZINE.
Cloudflare announced Clairic on October 19, 2026, an extraction model it calls a System One class tool. Instead of generating text word by word like an LLM, Clairic analyzes input in a single pass and outputs the requested values directly, including confidence levels and an 80% prediction range for estimates. It handles text and photographed images, returns null when information is absent, and computes derived figures such as outage duration.
Clairic answers in JSON, carries a confidence level on each extraction, and returns the value null when the requested field is not in the input, which Cloudflare says cuts the problem of filling in information that does not exist. In a worked example it read a customer failure inquiry, then calculated how long the failure had run and estimated affected users and lost orders in about 0.55 seconds.
Cloudflare classifies it as an extraction model, separate from the decision models Jev and Clef, which pick from prepared options. The company calls Clairic the first model to expand System One to information extraction and numerical prediction.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.