Cloudflare's Clairic Model Beats LLMs on Document Extraction

Cloudflare, the internet infrastructure company, has released a preview version of a new model called Clairic on Workers AI, its platform for running AI models. A regular large language model (LLM) generates an answer word by word. Clairic works differently. It pulls the needed values out of a document in a single pass. In a test built around processing an insurance claim, the Clairic-based pipeline finished in 19.3 seconds. A pipeline where the general-purpose GLM-5.3 model handled all three steps itself took 61.7 seconds. That makes Clairic roughly 3.2 times faster. Cloudflare also says a Clairic-based setup costs 49% less per million claims processed. The model returns fields such as invoice number, payment date, or damage amount. Each field comes with a confidence score attached.

The result points to a division of labor inside AI pipelines. Some steps just need a handful of fields pulled from a document. They do not need a full reasoning model for that. A narrow, specialized model handles this part faster and cheaper. That frees up the larger model for steps where reasoning actually matters. Cloudflare showed this gap on a routine task, processing an insurance claim, not on a benchmark built to impress. That makes these numbers easier to trust than the typical test a lab reports on itself.