Clairic

Clairic is Cloudflare’s first extraction model. Cloudflare announced it on 19 October 2026 as a preview on Workers AI, and calls it “a new class of System One-style models”. Cloudflare’s post credits TypeSafe AI with the System One idea. TypeSafe coined the term and had not commented on the new class as of 19 October 2026. Until now every System One model on this site has been a decision model: it picks from options you list. Clairic returns values instead, read from the input, computed from it, or predicted beyond it: an invoice total, a delivery date, the days a parcel is late, a repair cost. Like Clef, it answers in one parallel pass and generates no tokens. The extraction model guide covers the class.

You send a state (text, plus up to four images) and fields, a map of named fields. Each field has a type and optional plain-language instructions. Field type What comes back text A span of the input, with its offsets number A value read or computed, with integer, decimals and unit options date, time, datetime The value in ISO form duration A length of time in the unit you choose list Matching text, numbers or dates in input order, with spans estimate A distribution inside your minimum and maximum, on your step grid forecast One estimate per future step of a series Every scalar answer carries value, p_null (the probability the value is not in the input), confidence and guess. When the value is missing, value is null and guess shows what the model would otherwise have picked. Estimates add a median, an 80% interval from p10 to p90, and p_below and p_above; all 99 percentiles are available on request. Forecasts return the same numbers as arrays, one per step. The API is not Jev-compatible. It takes fields, not questions, and answers under answers. The Clairic field types guide covers each type, its settings and what its answer holds.

There are two ways in. From a Worker, call env.AI.run("@cf/cloudflare/clairic", {...}) through a Workers AI binding. From anywhere else, send a POST to https://api.cloudflare.com/client/v4/accounts/{account_id}/ai/run/@cf/cloudflare/clairic with a Cloudflare API token as a bearer token. The body is the same either way: model: "clairic", a state and at least one field. The Cloudflare integration page covers Workers AI setup for Clairic, Clef and Jev. The invoice example from the Workers AI model page, cut to four of its six fields, called from a Worker: const response = await env.AI.run("@cf/cloudflare/clairic", { model: "clairic", state: "Invoice INV-2291 from Northwind Supply, issued 2026-09-02, net 30. 4 pallets at $250.00 each, shipping $80.00. Contact: Dana Ruiz, Priya Shah.", fields: { invoice_number: { type: "text" }, due_date: { type: "date", instructions: "Issue date plus the net payment days" }, total: { type: "number", decimals: 2, unit: "USD", instructions: "Amount due including shipping" }, po_number: { type: "text", instructions: "Purchase order number" }, }, }); The answers block of the page’s response, for those four fields: { "invoice_number": {"type": "text", "p_null": 3.8e-10, "span": [8, 16], "guess": "INV-2291", "confidence": 1, "value": "INV-2291"}, "due_date": {"type": "date", "p_null": 1.9e-08, "guess": "2026-10-02", "confidence": 0.9999, "value": "2026-10-02"}, "total": {"type": "number", "p_null": 2.3e-08, "guess": 1080, "confidence": 0.9999, "value": 1080}, "po_number": {"type": "text", "p_null": 1, "span": [8, 16], "guess": "INV-2291", "confidence": 0.9995, "value": null} } The due date and the total appear nowhere in the text. Clairic computed both. The invoice has no PO number, so po_number comes back null with a p_null of 1. Each field can also set a null_threshold, the p_null above which its value comes back null (0.5 by default). Number fie

Every figure here is vendor-run. No independent board has measured Clairic as of 19 October 2026. On 21 next-day forecasts of US airport checkpoint counts, the mean error was 3.3%, and 17 of the 21 actual counts fell inside the 80% interval. In a claims pipeline, Clairic then Clef then GLM-5.3 took 19.3 s end to end, against 61.7 s with GLM-5.3 at every step. The extract step took 0.46 s against 16.1 s. Cost per million claims was $6,032 against $11,740. The blog’s table calls that about 3.2x faster and 49% cheaper. Co-author Michelle Chen’s X thread says Cloudflare’s live demo shows the cascade 3.5x faster and 77% cheaper. The two sets of figures disagree. The demo runs live and lets you pick the LLM and its reasoning level, so its figures depend on that setup. Neither set is independent. Cloudflare says both pipelines got the facts and the decision right in every run, and Clairic’s repair estimate was the same each time.

Clairic suits structured-output calls that repeat the same shape: totals, dates and durations from emails, invoices and receipt photos, including values one or two steps of arithmetic away. Cloudflare says thirty fields cost about the same as three. It also fits in front of a decision model: Clairic extracts the values, then Clef decides. Clef vs Clairic and Jev vs Clairic cover when to use which and how to chain them. Clairic vs GLiNER sets it beside the open-weights span extractor, and Clairic vs LLM structured outputs beside JSON mode on GPT, Claude and Gemini.

Cloudflare lists the preview’s limits itself. Longer chains of dependent steps and totals across many table rows are not reliable, so “have Clairic predict the value and let code do the math.” Estimates need evidence in the input. “Calibration is measured, not guaranteed,” and intervals can run narrow on unfamiliar data. Cloudflare calls forecasting “basic” and says “the architecture makes every answer well formed, not necessarily correct.” It writes no prose and does not pick from options.

Clairic is in preview on Workers AI as @cf/cloudflare/clairic, at $0.10 per million input tokens with no output charge. Responses report 0 output tokens. Input counts the state, any images and the field definitions. In the docs’ invoice example the field definitions took 472 of 655 input tokens, so a long schema can cost more than the text it reads. Cloudflare’s terms for the model say it “may cease providing it”. Cloudflare says open weights are coming soon. Michelle Chen’s thread says the team wants to make “a few tweaks” first. Neither has given a date or licence as of 19 October 2026.

Field typesText, Number, Date, Time, Datetime, Duration, List, Estimate, Forecast Total context65,536 tokens Rate limitNot published on the Workers AI model page as of 19 October 2026. EndpointPOST https://api.cloudflare.com/client/v4/accounts/{account_id}/ai/run/@cf/cloudflare/clairic, or env.AI.run("@cf/cloudflare/clairic", {...}) from a Worker LicenceNot published as of 20 Oct 2026 SDKs Workers AI lists a 65,536-token context window, checked 2026-10-19. There is no cap on the number of fields: fields that do not fit beside the state in one 8,192-token pass are split across parallel passes that each see the whole input, and the answers are merged. The usage block reports the number of passes. A request takes up to 4 images (PNG, JPEG or WebP, 4 MiB each, 8 MiB in total decoded) inside a 16 MiB request body; remote URLs are not accepted, and images are placed before the state. An optional reference_date (an ISO date) anchors dates written without a year. The response carries a truncated flag.

@cf/cloudflare/clairic, 19 Oct 2026, The preview, which Cloudflare calls an early checkpoint. Cloudflare says it intends to release the weights openly soon; co-author Michelle Chen says that comes after "a few tweaks". No date or licence published as of 19 October 2026.

What people use Clairic for, one page per pattern. Data and operationsStarter

There are two routes. Clairic, Cloudflare's extraction model, takes named fields and returns typed values, each with the probability that it is missing. Jev cannot write values, so code finds candidate spans, a choice question picks one, and code copies it unchanged. Choice NoulTextNumberDate

The most-starred and most-viewed entries in the directory. All Clairic examples.