Cloudflare previews Clairic, a typed data extraction model on Workers AI Other

Cloudflare has released Clairic, a model that pulls typed values out of text and images instead of generating text. It is in preview on Workers AI as @cf/cloudflare/clairic, callable from a Worker or through the REST API.

Cloudflare announced Clairic on October 19, 2026. It calls it the first "System One extraction model": it answers every requested field in a single parallel pass, without reasoning and without generating tokens. Cloudflare says it typically responds in under a second. The company intends to release the weights openly but has not given a date.

What it is

• Input. A request is a state (the text to read) plus up to four optional images (PNG, JPEG or WebP), and a map of typed fields. Each field gets a type and optional plain-language instructions.
• Extracted text. Text answers are spans copied from the input, returned with their character offsets.
• Computed values. Numbers, dates, times, durations and lists, read from the input or worked out from it. Cloudflare says one- and two-step calculations such as totals, differences and date arithmetic seem reliable. Longer chains of steps are not yet.
• Estimates and forecasts. Values the input does not state come back as a distribution inside bounds you set, with a median and an 80% interval. A forecast field returns one estimate per future step, each with its own 80% interval.
• Typed output. Every answer is a value of the declared type, or null when the input does not contain it. Scalar answers come with a confidence score and p_null, the probability that the value is missing.
• Use cases. Cloudflare's examples include invoices, a support email, a photo of a receipt, a home price estimate and a daily time series.
• Limits. Context window of 65,536 tokens. Fields that do not fit next to the input in one 8,192-token pass are split across parallel passes. There is no upper limit on the number of fields.

Price

Cloudflare's own tests

Cloudflare ran a damaged-product claim pipeline (extract, decide, reply) two ways: Clairic plus its Clef decision model plus GLM-5.3, and GLM-5.3 for every step.

That is 49% lower cost and about 3.2 times faster. In a separate test on 21 days of U.S. TSA checkpoint counts, Clairic's next-day forecasts were off by 3.3% on average, and 17 of 21 actual counts fell inside its 80% interval. Cloudflare says forecasting is basic and is not meant to replace dedicated time-series models.

Sources

• Cloudflare Blog: Introducing Clairic, a new System One class of extraction models
• Cloudflare Workers AI docs: clairic model page