Published
Updated
Cloudflare posts Clairic, a System One extraction model at $0.10 per million input
This topic was created 11 days ago, and the information it contains may have evolved or changed since then.
Cloudflare posted Clairic on October 19, 2026. The blog's cascade ran in 19.3 s against 61.7 s for GLM-5.3 at every step, $6,032 against $11,740 per million claims. Docs price input at $0.10 per million tokens. This desk did not call the API.
Michelle Chen posted Clairic on October 19, 2026, at 17:54 UTC. The text calls it a system one model for extracting, computing, estimating, and predicting values, and links the Cloudflare blog. The blog’s schema.org block says datePublished 2026-10-19T17:48:26.614Z. The byline is Michelle Chen.
Kevin Flansburg, 18:24 UTC, calls it a system one model based on regression instead of classification, names decision models like Clef as the other side, and says open weights are coming soon. The share image on this page is the still from Michelle’s 22:32 UTC post, status 2112310950309355943.
The hosted id is @cf/cloudflare/clairic on Workers AI. A request is a state, up to four images, and a map of typed fields. The blog names text, number, date, time, datetime, duration, list, estimate, and forecast. Each scalar answer carries p_null and confidence. Estimate answers carry a median and an 80% interval. A missing field returns null. The page says the model runs one forward pass and does not generate output tokens.
The damaged-claim table on that blog runs Clairic, then Clef, then GLM-5.3, against GLM-5.3 at every step. End to end is 19.3 s against 61.7 s. Extract is 0.46 s against 16.1 s. Decide is 0.22 s against 21.9 s. Reply is 18.6 s against 16.4 s. Cost per million claims is $6,032 against $11,740. The prose under the table writes 69% less time, about 3.2 times faster, and 49% cheaper. Michelle’s 17:54 UTC follow-up prints 3.5x faster and 77% cheaper for a live demo. Those two prints stay on their own rows.
A TSA checkpoint series on the same blog, 21 one-day-ahead forecasts, is off by 3.3% on average, with 17 of 21 actual counts inside the 80% interval. The page says three of the four misses fall on the Columbus Day weekend.
Workers AI docs, opened October 20, list a 65,536-token window, vision, preview, and $0.10 per million input tokens. REST is POST https://api.cloudflare.com/client/v4/accounts/{account}/ai/run/@cf/cloudflare/clairic. We did not call it.
Clef stays on its own page. Clef-omni stays on Clef-omni. Clairic is an extraction model. It does not pick among a list of options. Other hosted decision calls are on decision models like Jev.
This site's reading
Editorial notes evaluating claims against primary sources, contextualizing findings alongside related implementations, and defining technical terms.
Verify
Michelle Chen, @michellechen, October 19, 2026, 17:54 UTC, status 2112241006104244584. The text introduces Clairic as a system one model for extracting, computing, estimating, and predicting values from unstructured inputs, and links https://blog.cloudflare.com/clairic-extraction-model/. The 17:54 UTC follow-up, status 2112241010818605460, says a cascaded pipeline with Clairic, Clef, and an LLM is 3.5x faster and 77% cheaper than using an LLM alone. Kevin Flansburg, @kevin_flansburg, 18:24 UTC, status 2112248553703928141, calls Clairic a system one model based on regression instead of classification, names decision models like Clef as the classification side, and says open weights are coming soon. The share image on this page is the still from Michelle's 22:32 UTC post, status 2112310950309355943.
The blog schema.org block says datePublished 2026-10-19T17:48:26.614Z. The byline is Michelle Chen. Hosted id @cf/cloudflare/clairic on Workers AI. Open weights are intended later. A request is a state, up to four images, and a map of typed fields. Types named on the page are text, number, date, time, datetime, duration, list, estimate, and forecast. Each scalar answer carries p_null and confidence. Estimate answers carry a median and an 80% interval.
The damaged-claim cascade table, Clairic then Clef then GLM-5.3 against GLM-5.3 at every step: 19.3 s against 61.7 s end to end. Extract 0.46 s against 16.1 s. Decide 0.22 s against 21.9 s. Reply 18.6 s against 16.4 s. Cost per million claims $6,032 against $11,740. The prose under that table writes 69% less time, about 3.2 times faster, and 49% cheaper. TSA checkpoint forecasts, 21 days: 3.3% average error, 17 of 21 actual counts inside the 80% interval.
Workers AI docs, opened October 20. Context window 65,536 tokens. Vision yes. Unit pricing $0.10 per million input tokens. Preview. REST is POST https://api.cloudflare.com/client/v4/accounts/{account}/ai/run/@cf/cloudflare/clairic. We did not call the endpoint.
Compare
19.3 s against 61.7 s, and $6,032 against $11,740 per million claims, are the blog table for one damaged-claim pipeline. Michelle's 17:54 UTC post prints 3.5x faster and 77% cheaper for a live demo. Those two prints are different. 3.2 times and 49% cheaper are arithmetic on the table.
0.46 s is Clairic's extract step on that table. 0.22 s is Clef's decide step. Clef's October 1 median of 209.3 ms stays on the Clef page.
$0.10 per million input tokens is the Clairic card. Clef stays at $0.24. Clef-flash, opened October 9, is $0.038. Those three prices are three cards.
Clairic returns typed values from a field schema. It does not pick among a list of options. Clef and Jev stay the decision models on this desk.
Terms
- Clairic
- Cloudflare's System One extraction model, preview on Workers AI as @cf/cloudflare/clairic, posted October 19, 2026. A request is a state plus typed fields. The reply is a value of the declared type, or null, with confidence and p_null.
- 19.3 s / 61.7 s
- Cloudflare's damaged-claim cascade, Clairic then Clef then GLM-5.3, against GLM-5.3 at every step. Cost on that table is $6,032 against $11,740 per million claims.
- $0.10
- Clairic input price on the Workers AI card, per million tokens. The card lists no output charge.