AI News Feed
Market watch
South Korea Navy plans drone carrier, nuclear submarines by early 2040s to counter North KoreaReview Details BenQ MA270S 5K Monitor for MacBook Pro Dual-Screen SetupTechRadar guide maps NBA 2026/27 viewing options as Knicks begin title defenseiQIYI’s 580-Day AI Push: From ‘Skip Watch’ to Script and Image WorkshopsUnverified Kim Jong-un death rumors spread overseas; state media reports public visitFive Degrees May Decide Whether Outdoor Networks Survive Rising Heat, Extreme Networks Executive Says
South Korea Navy plans drone carrier, nuclear submarines by early 2040s to counter North KoreaReview Details BenQ MA270S 5K Monitor for MacBook Pro Dual-Screen SetupTechRadar guide maps NBA 2026/27 viewing options as Knicks begin title defenseiQIYI’s 580-Day AI Push: From ‘Skip Watch’ to Script and Image WorkshopsUnverified Kim Jong-un death rumors spread overseas; state media reports public visitFive Degrees May Decide Whether Outdoor Networks Survive Rising Heat, Extreme Networks Executive Says
Products & Applications

Cloudflare Releases Clairic Extraction Model Preview on Workers AI

Cloudflare previewed Clairic on Workers AI, an extraction model returning typed values in one pass; a test claims pipeline fell to 19.3 seconds from 61.7.

AI News Feed · 2026-10-20 17:40

Clairic runs on Cloudflare-hosted GPUs through Workers AI at @cf/cloudflare/clairic. Cloudflare has not disclosed the parameter count, but it lists a 65,536-token context window and support for up to four images per request. There is no local option until weights ship; Cloudflare has promised open weights but they were not yet available, according to MarkTechPost. In Cloudflare's tests, extraction was sub-second, but no standard accuracy benchmark has been published. Cloudflare's best claim was a 0.46-second extraction step against 16.1 seconds for GLM-5.3 in its claims pipeline; its worst included four of 21 TSA next-day forecasts outside the model's 80% interval, and the report said multi-step arithmetic is unreliable.

Clairic builds on the System One idea from TypeSafe's Jev, named for Daniel Kahneman's fast, intuitive thinking. Jev and Cloudflare's own Clef are decision models that score options a user lists in advance. Clairic covers answers that cannot be listed, such as an invoice total, a date or a repair cost. Cloudflare describes a three-part stack: extraction models turn raw input into typed values, decision models turn values into decisions, and LLMs turn decisions into words.

A Clairic request contains a state, meaning text plus up to four images, and a map of named, typed fields. Each field can carry plain-language instructions. Supported types are text, number, date, time, datetime, duration, list, estimate and forecast, according to the Workers AI changelog. Every scalar answer includes p_null, the probability that the value is absent, plus a confidence score. Text answers return exact spans with character offsets. Estimates and forecasts return a median and an 80% interval from p10 to p90, and forecasts cover up to 64 steps. There is no field limit. Fields that do not fit beside the input in one 8,192-token pass split into parallel passes, each of which sees the whole input. In a documentation example, a 61-token invoice ran in three passes and billed 655 input tokens, including 472 schema tokens; Cloudflare noted that wide schemas can cost more than the document itself.

Cloudflare's examples included a support email in which Clairic added price and tax into $701.10 and computed 11 days since delivery, neither of which appeared in the email. A missing tracking number returned null instead of a plausible guess. On a skewed receipt photo, it read all seven present values and returned null for an absent discount; its lowest confidence was 0.776, on the order number. For forecasting, Cloudflare predicted daily TSA checkpoint counts for 21 days. Average error was 3.3%, and 17 of 21 actual counts fell inside the 80% interval. Three of the four misses fell on Columbus Day weekend.

The main test was a damaged-product claims pipeline. Clairic extracted facts, Clef chose refund or repair, and GLM-5.3 wrote the reply. Cloudflare compared that cascade with GLM-5.3 handling all three steps. The cascade finished in 19.3 seconds versus 61.7 seconds, about 3.2 times faster. Extraction took 0.46 seconds against 16.1 seconds, and the decision step took 0.22 seconds against 21.9 seconds. Cost per million claims fell 49%, from $11,740 to $6,032. The reply step still dominated at 18.6 of the 19.3 seconds. Both pipelines got the facts and decision right. Clairic's repair estimate was identical across runs, while GLM-5.3 returned $250, $250 and $220. Cloudflare said Gemma 4 and Qwen failed the all-LLM version.

Clairic is priced at $0.10 per million input tokens with no output charge, according to MarkTechPost. That undercuts Clef's $0.24 per million input tokens but sits above Jev's $0.042, with free output. In the comparison published by the report, GLM-5.3 costs $1.40 per million input tokens and $4.40 per million output tokens. Clef has 27 billion parameters, a 65,536-token context window and image input, and is available on Workers AI with Apache 2.0 weights. Jev is available through TypeSafe's early-access API. GLM-5.3, described as a System Two LLM, has a 1,048,576-token context window. Clairic is live as a preview at @cf/cloudflare/clairic and can be called through the Workers AI binding with env.AI.run() or the REST API; AI Gateway works with the same endpoints.