Thursday, November 19, 2026
Translate
PrivacySubmit newsShare your story Newsletter Advertise
Breaking
AI news

Cloudflare Makes Its Clairic AI Extraction Models Open-Weight on Hugging Face

Cloudflare has released open-weight versions of its Clairic and Clairic-flash AI extraction models on Workers AI and Hugging Face.

Cloudflare Makes Its Clairic AI Extraction Models Open-Weight on Hugging Face
Image: Cloudflare

Cloudflare announced on November 18, 2026, the release of open weights for an updated version of its Clairic model alongside a new 9-billion-parameter model called Clairic-flash. Both models focus on structured data extraction and are available on Cloudflare's Workers AI platform as well as Hugging Face.

Redesigned architecture for document extraction

Cloudflare stated that following preview feedback, it redesigned the Clairic model architecture to focus on extracting unstructured data from complex documents, including images and multi-page PDFs. For ordinary text and number extraction fields, the model now uses parallel slot autoregressive generation instead of a single forward pass.

The company also trained dedicated heads to process estimates, forecasts, durations, calculated dates, and explicit arithmetic. Cloudflare said the API maintains typed interface features, including confidence scores, character spans, and probabilities.

The TechUpscale Brief. The day's cyber, AI and tech news in one short email, every weekday morning. Free. Unsubscribe anytime.

Image reading capabilities and latency

The updated Clairic model family allows users to extract data from up to four images or PDF pages in a single request. Cloudflare reported that on its internal benchmarks covering text and images, image reading errors dropped from 207 to zero.

In complete HTTP response latency testing on a self-hosted server, Cloudflare reported that both Clairic and Clairic-flash delivered responses in under one second across all tested evaluation sets, including fresh text fields, image fields, commercial leases, loan pricing, and traffic forecasting.

Model endpoints and pricing structure

Cloudflare is hosting the 9B variant at the @cf/cloudflare/clairic-flash endpoint, while the 27B model resides at @cf/cloudflare/clairic. Full open weights for both variants have been published on Hugging Face.

The 27B Clairic model is priced at $0.24 per M input tokens, and Clairic-flash is priced at $0.10 per M input tokens. Cloudflare stated it only charges for input tokens and absorbs the cost of output token generation. Until December 1, 2026, Cloudflare is offering a promotional discount pricing both Clairic and Clairic-flash at $0.10 per M input tokens.

What to do

  • Access the 9B model variant at the @cf/cloudflare/clairic-flash endpoint on Workers AI, or the 27B model at @cf/cloudflare/clairic.
  • Download open weights for both Clairic and Clairic-flash directly from Hugging Face.
  • Take advantage of the $0.10 per M input tokens promotional rate on the 27B Clairic model before pricing changes on December 1, 2026.
Key facts and where they come from
  • Cloudflare announced two open-weight AI model releases, Clairic and Clairic-flash.
    Today, we’re announcing two open-weight model releases:
  • Clairic-flash is a 9-billion-parameter extraction and prediction model.
    delivers strong extraction and prediction performance in a compact, 9-billion-parameter model.
  • Both models are hosted on Workers AI and published with open weights on Hugging Face.
    Both extraction models are hosted on Workers AI and full open weights are available on Hugging Face.
  • Cloudflare offers promotional pricing of $0.10 per M input tokens for both models until December 1, 2026.
    until December 1, 2026, both Clairic and Clairic-flash will cost $0.10 per M input tokens.
  • Cloudflare only charges for input tokens and absorbs output token costs.
    we decided to keep the System One pricing principles and are only charging on input tokens: we eat the cost of the output token generation.

The catch

  • Cloudflare acknowledged that its benchmark results comparing Clairic against competitor models are self-reported internal evals.
  • The promotional rate of $0.10 per M input tokens for the 27B Clairic model ends on December 1, 2026, after which it increases to $0.24 per M input tokens.
  • Requests to the document extraction models are limited to a maximum of four images or PDF pages per request.

Add TechUpscale as a preferred source on Google

The TechUpscale Brief

The day's cyber, AI and tech news in one short email, every weekday morning. Free. Unsubscribe anytime.

I'm most interested in

Join the conversation

Comments are moderated. Be civil and stay on the story: no personal attacks, hate, spam or ads, and no one else's private details. Your email is never shown. Privacy policy.

More AI