Cloudflare previews Clairic extraction model on Workers AI
Cloudflare has introduced Clairic, a System One extraction model that returns values rather than text. It is available in preview on Workers AI, with open weights planned but no release date announced.

Key takeaways · 4
- 01
Clairic is in preview on Workers AI as `@cf/cloudflare/clairic`; Cloudflare has not given a date for its planned open-weights release.
- 02
The model accepts typed schemas, including text, numbers, dates, lists, estimates and forecasts, and can process up to four images with the input.
- 03
Cloudflare charges $0.10 per million input tokens and says there is no output-token charge.
- 04
Cloudflare says forecasts follow recent patterns but may miss events absent from the data, and Clairic is not a replacement for a time-series model trained on a user's data.
A schema-driven extraction model
Clairic takes an input state and a schema of typed fields, then returns one typed value for each field.[1] Supported types include text, number, date, time, datetime, duration, list, estimate and forecast.[1] Cloudflare says the model can extract values from the input, compute them, or predict values beyond it.[1] It reads the input in one pass and returns null when the input does not determine a requested value.[1] Cloudflare says there is no upper limit on requested fields: fields that do not fit alongside the input are split across parallel passes.[1]
Preview access and pricing
Clairic is available in preview through Workers AI under the identifier `@cf/cloudflare/clairic`.[1] It can be called through a Workers AI binding or the REST API, and can also be used with AI Gateway.[1] Cloudflare lists a context window of 65,536 tokens.[2] The listed price is $0.10 per million input tokens, with no charge for output tokens.[1] Cloudflare says it intends to release Clairic with open weights, but has not provided a release date.[1] Teams assessing it can therefore test the preview now, while treating the open-weights timing as unknown.
Signals for review and forecasting
Each scalar answer includes a probability that the input does not determine the value and a confidence score.[1] Estimates and forecasts include a median and an 80% interval, with all 99 percentiles available as an option.[1] Cloudflare says estimates work best when inputs contain evidence such as comparable examples, history or descriptive text; without it, intervals can be wide.[1] The company says forecasts follow recent patterns, flatten toward the average several steps ahead, and do not anticipate events absent from the series, such as holidays.[1] It cautions that Clairic is not a substitute for a time-series model trained on a user's data.[1]
Examples and reported results
In a Cloudflare support-email example, Clairic calculated a $701.10 total from a $649.00 price and $52.10 tax, and returned null for a tracking number missing from the email.[3] Cloudflare reported that its forecasts were off by 3.3% on average and that 17 of 21 actual counts fell within the model's 80% intervals.[3] For October 14, 2026, the company reported a median TSA screening-count prediction of 2,419,283 against an actual count of 2,419,474.[3] These are Cloudflare-reported examples and results; the company cautions that well-formed answers are not necessarily correct.[1]
Clairic offers teams a way to request structured values, calculations and forecasts directly, rather than asking a model to compose text that must then be parsed. Its null handling and uncertainty signals may help teams decide which outputs are suitable for automated workflows and which need review.
Why it matters
Put this to work — one session a day, built for your industry.
Create a free account for a daily session — eight questions and one real-work challenge, on the news that affects your role.
Start freeHow this developed
20 October 2026
Cloudflare previews Clairic extraction model on Workers AI