Cloudflare announced Clairic on October 19, 2026, releasing a preview of what it describes as the first model in a new class of System One extraction models, hosted on Workers AI with an open-weights release planned for later.
The model is built to pull typed values out of unstructured inputs, extracting, computing, estimating, and forecasting rather than generating text. Cloudflare says Clairic does this at higher accuracy, faster latency, and lower cost than traditional large language models.
Availability and Pricing on Workers AI
Clairic is available in preview as @cf/cloudflare/clairic, callable from a Cloudflare Worker with env.AI.run() or through the REST API. The official model documentation lists a 65,536-token context window, vision support, and unit pricing of $0.10 per million input tokens.
Because the model generates no output tokens, billing applies to input only. Cloudflare said it intends to release the weights openly after further checkpoint work, without giving a date.
How the Extraction Model Works
A request pairs a state, the text to read plus up to four optional images, with a map of typed fields. Supported field types are text, number, date, time, datetime, duration, list, estimate, and forecast, each with optional plain-language instructions. The model answers every field in a single parallel forward pass, with no reasoning step and no generated tokens, and Cloudflare says it typically responds in under a second. Fields that do not fit beside the input in one 8,192-token pass are split across parallel passes that each see the whole input, then merged. The documentation limits embedded images to four PNG, JPEG, or WebP files of up to 4 MiB each.
Every scalar answer carries p_null, the probability that the value is absent from the input, and a confidence score. Estimate fields return a distribution inside caller-declared bounds, with a median and an 80% interval from p10 to p90 and all 99 percentiles on request. A field the input does not determine returns null, with a guess field showing what the model would otherwise have picked, rather than a fabricated value.
Cloudflare places Clairic in the System One category alongside the decision models that preceded it. TypeSafe introduced its Jev decision model about a month before the announcement, and Cloudflare released its Clef and Clef-flash decision models on October 1, 2026, open-sourcing them on Hugging Face under an Apache 2.0 license; Clef was post-trained on Qwen base models and is Jev-API compatible. Where decision models classify an input among listed options and large language models generate open-ended text, Clairic returns typed values directly: spans, numbers, dates, lists, and distributions. The company describes the three categories as complementary stages, with extraction feeding decisions and decisions feeding the final generated reply.
Documented Examples and Reported Tests
In Cloudflare’s support-email example, the model computed a $701.10 total from a stated $649.00 price plus $52.10 tax in the same pass, derived 11 days between the September 28, 2026 delivery and the October 9, 2026 message date, listed cracked screen and dented corner as exact spans of the input, and returned null for a tracking number the email never contained. In a receipt-photograph example, Clairic extracted the order number, purchase datetime, quantity, unit price, subtotal, sales tax, and tip with confidence scores, and returned null for a discount the receipt did not show.
For estimation, the announcement shows Clairic pricing a home from comparable-sales text: the model returned a value of $418,000 with a median of $415,000 and an 80% interval of $370,000 to $470,000 inside caller-set bounds. For forecasting, Cloudflare reports a 21-day test in which Clairic predicted next-day TSA airport checkpoint screening counts from the public series. The company says the forecasts were off by 3.3% on average and that 17 of the 21 actual counts landed inside the model’s 80% interval, with three of the four misses falling on the Columbus Day weekend. For October 14, 2026, Cloudflare reported an actual count of 2,419,474 against a predicted median of 2,419,283.
Cloudflare also reports a pipeline comparison that pairs Clairic with Clef and the GLM-5.3 language model against an all-GLM-5.3 version of the same damaged-product claims workflow. According to the company, the cascade finished in 19.3 seconds against 61.7 seconds, 69% less time or about 3.2 times faster, and cost $6,032 per million claims against $11,740, or 49% less. Clairic and Clef handled extraction and the decision in under a second combined, where GLM-5.3 took 38 seconds for the same two steps, and Cloudflare says Gemma 4 and Qwen could not derive values such as repair cost and failed the task in the all-LLM version. Cloudflare built the workflow into a live demo.
Preview Limitations and Open-Weights Plans
Cloudflare calls the current endpoint an early checkpoint and lists several limitations. One- and two-step arithmetic such as totals, differences, conversions, and date arithmetic seem reliable, it said, but longer chains of dependent steps and aggregation across many table rows are hard for a one-pass model without a scratchpad; it advises users to rely on Clairic for the prediction and handle the arithmetic in code. Estimates need evidence in the input, calibration is measured rather than guaranteed and intervals can run narrow on unfamiliar data, and forecasting is basic rather than a substitute for dedicated time-series models.
The company advises using confidence scores, p_null, and the intervals to decide what to automate and what to route for additional review, because the model’s answers are guaranteed to be well formed, not to be right. Cloudflare said it will keep working on the arithmetic, estimation, and forecasting sides of the model with each checkpoint, and that it plans to release Clairic open-weights afterward. The name, it added, plays on clerical work, clarity, and clairvoyance, and doubles as shorthand for Cloudflare’s AI regression model.