Clairic extracts typed values without generating an answer token by token

Cloudflare’s Clairic preview extracts text spans, numbers, dates and other typed values from unstructured input in a single model pass. It can also estimate values and return uncertainty intervals, rather than produce an open-ended answer. Cloudflare distinguishes this extraction step from choosing an action or writing a response. Correct output formatting does not guarantee a correct value, especially with unfamiliar inputs.

Cloudflare · Oct 19, 2026

Why it matters

A narrow extraction step may be easier to check than asking a general chatbot to perform an entire workflow.

More context

The hosted model is available through Workers AI, using a Worker binding or the REST API. Try it on a labelled extraction task before automating decisions. Cloudflare describes the endpoint as an early checkpoint and recommends ordinary code for longer arithmetic chains.

Keep in mind

Open weights are promised, not released in this announcement. Confidence can be poorly calibrated on unfamiliar data; check service billing before scaling usage.

Original read