Cloudflare launches Clairic, a model that answers with values, not text
Clairic reads text and up to four images and returns typed fields, numbers, dates or forecasts with a confidence score, for $0.10 per million input tokens and no output charge.
Cloudflare added Clairic to Workers AI on 19 October, a model built for extracting structured values rather than generating text. Where Cloudflare’s existing Clef models pick among options you define up front, Clairic is for the cases where you cannot list the answers in advance, things like an invoice total, a delivery date or a repair cost.
You give it a schema of typed fields, text plus up to four images, and it returns one typed value per field: text, number, date, time, datetime, duration, a list, or an estimate or forecast with a median and an 80% interval. Every answer comes with a confidence score and a p_null probability, and it generates no tokens, just values, in a single pass. It’s called the same way as other Workers AI models, through env.AI.run() with the identifier @cf/cloudflare/clairic. Pricing is $0.10 per million input tokens, with no output charge since nothing is generated.
const result = await env.AI.run('@cf/cloudflare/clairic', { schema, input });Why it matters: extracting a handful of fields from a document or image today usually means prompting a general LLM and parsing its text reply, hoping the format holds. A model that returns typed fields directly, with a confidence score per field, removes a parsing step and gives you something to check before you trust the output.
The caveat: this is a first release in a new model class from Cloudflare, so there’s no independent benchmark yet for how well it extracts against ambiguous or messy input.