AI Aulis

AI NEWS IN ENGLISH · 20 November 2026

The most important AI news every morning. AI writes and checks the stories against the original sources – check important facts in the sources.

41 stories · next issue tomorrow morning · How the paper is made →

← Issue 20 November 2026Paper

Cloudflare releases Clairic and 9-billion-parameter Clairic-flash as open-weight models

AI-generated contentWritten by AI

blog.cloudflare.com, x.com, technobezz.com, developers.cloudflare.com · 5 sources · Sources published: 19 October–19 November 2026 · Some source publication dates are unknown

The models extract information from text, images and now also PDF files, and can be downloaded from Hugging Face. The existing Workers AI address now serves a larger 27-billion-parameter model, and its price rises from $0.10 to $0.24 per million input tokens on 1 December.

Cloudflare announced the update in a blog post on 18 November. Clairic is an extraction model. The user gives it text, an image or a PDF along with a list of typed fields, such as an invoice number, a due date and a total. The model returns a value and a confidence score for most fields. For text fields, it also shows where in the input it read the value.

The preview released in October read the input in a single forward pass and generated no tokens. The new version writes text and number fields token by token. Each field progresses on its own, but all fields are processed at the same time. Cloudflare said accuracy improved and latency increased slightly. An X post shared by Hugging Face still describes the model as working in a single forward pass, which matches the description of the preview.

The table shows results from Cloudflare’s own tests of the Clairic models and two competitors. Apple’s Vision text recognition read the text before Liquid AI’s D1.2 processed it. The other competitor is Scaledown’s Extract.

Test Clairic preview Clairic Clairic-flash D1.2 + OCR Scaledown
Fresh text fields, 96 requests 295/336 335/336 328/336 304/336 259/336
Mixed-format fields, 36 requests 207/226 221/226 213/226 116/226 156/226
Customer checks, 139 request types 1,033/1,263 1,263/1,263 1,263/1,263 973/1,263 653/1,263
Synthetic receipts in text and images 369/576 576/576 576/576 397/576 280/576

Source: Cloudflare, internal and self-reported tests.

Cloudflare said there are no public extraction benchmarks yet, and it advises treating the results as indicative only because they are self-reported. The company said errors in reading images fell from 207 to zero in internal tests. The figure matches the synthetic receipts row, where the preview got 369/576 right, making 207 errors.

Measuring on its own server, Cloudflare recorded a median latency for the full HTTP response on text fields of 358 milliseconds for Clairic and 155 milliseconds for Clairic-flash.

Cloudflare bills only for input tokens, but the billed count can be several times the length of the document. The Workers AI documentation says each text field gets its own pass, and every pass is billed for the full text and all images. In the documentation’s example, a 61-token invoice and the field definitions produced 655 billable tokens over three passes.

The preview’s address, @cf/cloudflare/clairic, now automatically serves the 27-billion-parameter model. From the beginning of December, its price will be $0.24 per million input tokens. The cheaper 9-billion-parameter model now has the address @cf/cloudflare/clairic-flash, and its price stays at $0.10. Cloudflare said the API schemas are unchanged.

A single request can include up to four images or PDF pages in total. Each image or PDF file can be up to 4 MiB, and all files together up to 8 MiB. Because the weights are open, users can download the model from Hugging Face and run it on their own hardware.

Report an error in this story