Cloudflare on 19 October 2026 announced a preview of Clairic, which it calls an extraction model: it reads unstructured text and images and returns typed values such as numbers, dates and lists, without generating text. In a test Cloudflare ran itself, a pipeline using Clairic cost 49% less and took 69% less time than the same pipeline run entirely on an LLM.
Clairic is hosted on Cloudflare’s Workers AI at $0.10 per million input tokens, according to its model page. Cloudflare says it intends to release the model with open weights soon.
Business Pill · MODEL ROUTING
A one-minute explainer of model routing. It teaches the general idea only and says nothing about Cloudflare or Clairic.
The key insight: As we read it, Cloudflare is pricing Clairic on input tokens alone, which matches its claim that the model generates no output tokens, the part of an LLM call that cost the most in its own test.
Cloudflare’s post says Clairic can “extract, compute, estimate, and forecast values from unstructured inputs.” A request carries the text to read, plus up to four optional images, and a list of named fields, each with a type such as text, number, date, duration, list, estimate or forecast.
The model answers every field in a single pass. Cloudflare says it “only processes input tokens with a single forward pass, and thus has no charge for output tokens.” Each answer comes with a confidence score and a probability that the value is missing, and estimates come with a median and an 80% interval.
Cloudflare calls this a “System One” model, borrowing Daniel Kahneman’s term for fast thinking. It credits the idea to TypeSafe, which introduced a decision model called Jev a month earlier, and places Clairic alongside its own decision model, Clef.
Clairic by Cloudflare’s own numbers: the extraction step took 0.46 seconds against 16.1 seconds for GLM-5.3; the cascaded pipeline took 19.3 seconds against 61.7 seconds (69% less time) and cost $6,032 against $11,740 per million claims (49% less), in one test Cloudflare ran; list price $0.10 per million input tokens; a 21-day forecast example missed by 3.3% on average. Sources: Cloudflare blog (19 Oct 2026) and Workers AI model page.
What Was Posted on X
Introducing Clairic, from Cloudflare. A System One extraction modeI, capable of extracting, computing, estimating, and forecasting data from unstructured inputs. https://cfl.re/4zIgfpH
Cloudflare announced Clairic on X at 18:06 UTC on 19 October 2026, calling it a System One extraction model for extracting, computing, estimating and forecasting data from unstructured inputs. The speed and cost figures in this piece come from the blog post the announcement links to.
Today we introduced a novel system one model, Clairic, based on regression instead of classification (i.e. decision models like Clef). This enables estimation of continuous values without having to exhaustively list choices upfront. It also enables very efficient extraction of structured data from unstructured inputs. Open weights coming soon!
Kevin Flansburg wrote on X that Clairic is based on regression instead of classification, unlike decision models such as Clef, so it can estimate continuous values without listing the choices upfront. He added that open weights are coming soon; Cloudflare’s blog gives no date.
The Numbers in Cloudflare’s Test
Cloudflare ran one pipeline on a customer damage-claim email: extract the facts, decide what to do, then write a reply. In one version Clairic extracted, Clef decided and GLM-5.3 wrote the reply. In the other, GLM-5.3 did all three steps.
By Cloudflare’s figures, the extraction step took 0.46 seconds with Clairic against 16.1 seconds with GLM-5.3. The whole pipeline took 19.3 seconds against 61.7 seconds, which Cloudflare puts at 69% less time. It costed the two at $6,032 and $11,740 per million claims, or 49% less.
In a separate example, Cloudflare says Clairic’s forecasts over 21 days were off by 3.3% on average, and 17 of the 21 actual counts landed inside its 80% interval.
What Cloudflare Says It Cannot Do Yet
Cloudflare calls the hosted endpoint “an early checkpoint”. Its post says “Reading is strong, but multi-step arithmetic isn’t yet” and advises users to have Clairic predict the value and let code do the math.
It also says that “Forecasting is basic”, that the model is not designed to replace dedicated time-series models, and that its intervals “can run narrow on unfamiliar data.”
Connecting the Coverage
Clairic extends a line of Cloudflare modelpricing we have been tracking. On 10 October we reported that Cloudflare cut its Clef-flash decision model to $0.038 per million input tokens and shrank its context window from 64k to 24k tokens. Clairic’s list price of $0.10 per million input tokens sits above that, with a 65,536-token context window on its model page.
As we read it, the test splits the work three ways: Clairic extracted the facts, Clef chose the action and GLM-5.3 wrote the reply, so only one step generated text.
As we read it, output tokens drove the cost in Cloudflare’s table: it priced them at $5,848 per million claims in the cascaded pipeline and $11,194 in the all-LLM one.
As we read it, Cloudflare frames Clairic as one of a family of single-pass models, next to its Clef decision model and TypeSafe’s Jev.
Cloudflare blog, 19 October 2026
“Extraction models turn raw input into typed values, decision models turn values into decisions, and LLMs turn decisions into words”
Three Implications
PRICED ON INPUT ONLY Cloudflare’s model page lists Clairic at $0.10 per million input tokens.
ONE TEST, ONE PIPELINE The 49% cost cut and 69% time cut come from a damage-claim pipeline Cloudflare designed and ran.
A PREVIEW, NOT A RELEASE Cloudflare says the hosted endpoint is an early checkpoint and that open weights will follow, without a date.
The framework says: “Inference is a revenue engine: continuous, distributed across millions of applications and agents, incurring marginal cost with every query.”
As we read it, Clairic is aimed at that marginal cost: Cloudflare says asking for thirty fields costs about the same as asking for three, because every field is answered in the same pass. The framework was published in March 2026 and does not discuss Clairic.
What Is Not Established
The speed and cost comparison comes from a single pipeline that Cloudflare designed and ran, and Cloudflare has not published an independent benchmark. The open-weights release has no date, and the model is a preview whose behaviour Cloudflare expects to change with each checkpoint.
Business Engineer Framework
The State of The Inference Economy
A Business Engineer framework on why inference, not training, now drives the cost and revenue of AI.
Cloudflare is pitching Clairic as a cheaper, faster way to handle the structured-extraction calls that often go to LLMs, priced only on input tokens. The 49% saving and 69% cut in time come from one test Cloudflare ran, and Cloudflare itself flags multi-step arithmetic and forecasting as weak spots.
A note on sourcing. This piece is built on Cloudflare’s blog post of 19 October 2026, read in full, and its Workers AI model page for Clairic, read on 20 October 2026, plus our own earlier pieces on Cloudflare. The speed and cost figures are Cloudflare’s, from a test it designed and ran; we have not run the model. Nothing here is financial or investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
