Key Notes
- Cloudflare has published the weights for 27B Clairic and 9B Clairic-flash, two models for extracting structured fields from text, images and PDFs.
- Each request accepts up to four images or PDF pages in total, with typed answers and scoring information to support downstream processing.
- Hosted input pricing is ten cents per million tokens for both models until December 1, when Clairic rises to 24 cents while Flash stays at ten cents, with no output-token charge.
Cloudflare has released the weights for two AI models designed to turn documents into structured data, offering businesses an alternative to using a general-purpose chatbot for routine extraction. The updated Clairic and the smaller Clairic-flash can read text, images and PDFs and return fields such as supplier names, invoice totals and itemized purchases.
The company announced the releases on November 18. Both are available through Workers AI at an introductory price of $0.10 per million input tokens, with no charge for output tokens. The larger model’s price rises on December 1.
Two Models With Downloadable Weights
Clairic is a 27-billion-parameter model, while Clairic-flash has 9 billion parameters and targets faster, less expensive processing. Cloudflare has published both on Hugging Face under the Apache 2.0 license. The model card describes Clairic as a multimodal system that takes source material and a specification of the fields an application wants back.
The downloadable releases give teams the option to run the models on their own infrastructure, alongside Cloudflare’s hosted service. The Flash release includes the same style of typed outputs, with support for extraction, numerical estimates and forecasts.
From a Four-Page Invoice to Usable Records
An application can ask for a supplier, a list of purchased items, quantities, prices and a total in one request, even when the information spans several pages. The current limit is four images or rendered PDF pages combined per request. A four-page PDF therefore uses the entire page allowance.
The developer documentation lists a 65,536-token context window. Developers define the required output fields and their types, including text, numbers, dates, objects and lists. This gives downstream software a predictable structure for processing the result.
The model cards also describe confidence scores, null probabilities and completeness indicators. Where text evidence has a unique match, the response can include character positions linking an answer to the source. Image-only evidence does not provide those text spans, and model confidence still needs to be checked against an application’s own acceptance criteria.
The Ten-Cent Price Changes on December 1
Clairic’s hosted promotional rate ends on December 1, when its input price increases to $0.24 per million tokens. Clairic-flash remains at $0.10. Cloudflare continues to absorb output-token costs for both models.
There is a migration detail for existing preview users: the original Clairic endpoint is being upgraded from the smaller model to the new 27B version. Applications that retain that endpoint will get the upgraded model and its higher standard price after the promotion. Teams seeking the smaller model’s ongoing rate need to use the Clairic-flash endpoint.
The advertised input rate also needs to be read alongside the billing rules. Cloudflare explains that media is converted into input tokens and source material is billed again when a request needs multiple processing passes. One API request can therefore involve more than one billable pass through the same document.
Speed Claims Come From Cloudflare’s Own Tests
In Cloudflare’s fresh-text evaluation, median complete HTTP response time was 358 milliseconds for Clairic and 155 milliseconds for Clairic-flash. Those measurements came from its self-hosted test server and a specific set of 96 requests per model. They do not establish response times for every PDF or production deployment.
The release extends the focused-model strategy behind Clef, Cloudflare’s earlier family for selecting predefined answers and returning probabilities. Clairic addresses another recurring business task: getting information out of existing records. For an invoice-processing team, the useful comparison is how many fields are correct and how many documents still need a person to review them.
Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.