Subscribe
AI Models

Cloudflare ships open-weight Clairic models that read documents like a person

Thursday, November 19, 2026 · 5 min read

Cloudflare released Clairic and Clairic-flash as open-weight models under Apache-2.0, priced from $0.10 per million input tokens — and says image-reading errors on its own tests fell from 207 to zero.

Bullish$NET

📊 Scorecard: our Bullish call on $NET is a hit so far: from $300.91 to $303.68 at the latest close (+0.9%). Against QQQ the call is right: the stock is 3.1 points ahead of the index.

Cloudflare ships open-weight Clairic models that read documents like a person

The News

Cloudflare announced on Wednesday, November 18, that its updated document-extraction model Clairic (27 billion parameters) and a new compact Clairic-flash (9 billion parameters) are now open-weight under the Apache-2.0 licence on Hugging Face, hosted on Workers AI. The release re-introduces Cloudflare's frontier research group as the Workers AI superintelligence team, behind Clef and Clairic, with more models in the works.

The redesigned Clairic reads text, images and PDFs with better handling of tables, lists and dates, and grounds answers in source material — meaning it points back to where in the document an answer came from. Cloudflare says image-reading errors on its internal evaluations fell from 207 to 0, while admitting there are no public extraction benchmarks yet, so that figure is its own measurement, not an independent one.

Clairic is priced at $0.24 per million input tokens and Clairic-flash at $0.10, with both at a promotional $0.10 until December 1; only input tokens are charged. Cloudflare shares closed Wednesday at $300.91, up 5.66%.

Document extraction is the plumbing of AI — boring, essential, and now Cloudflare's to sell.

Why It Matters

The release matters because document extraction is the bottleneck for most real-world AI agents. An agent that books travel or processes invoices is only as good as its ability to read the paperwork — and a model that misreads a table or a date can cost real money. A small, cheap, open model that does this well would be used everywhere, from finance to healthcare to logistics.

The open-weight choice matters too. By publishing the model weights under Apache-2.0, Cloudflare invites developers to run Clairic anywhere, not just on its platform — which builds trust and adoption even if it gives up some lock-in. It is a bet that the money is in hosting and serving the model at scale, not in keeping it secret.

The new team name is the strategic tell. Re-introducing the research group as the Workers AI superintelligence team signals that Cloudflare intends to keep shipping frontier models, not just one release. Clef came before Clairic; more models are promised. If that pipeline delivers, Cloudflare's AI business becomes a story in its own right rather than an add-on to its networking products. For the share price, the context is a stock that has already run hard — NET closed Wednesday at $300.91, up 5.66% on the day — and the open-weight release gives the rally a product story to lean on rather than just momentum. If developers adopt Clairic as the default document reader for their agents, Workers AI gains a sticky workload that is hard to move elsewhere.

Analyst's Take

Our call is Bullish. Open-weight models that read documents accurately are exactly the kind of infrastructure developers need to build AI agents that handle real paperwork, and Cloudflare is selling that capability through its own Workers AI platform. The 5.66% jump in the shares on Wednesday suggests investors liked the combination: a useful model, a clear price, and a new research team with more to come.

The caveat is the 207-to-0 figure. It comes from Cloudflare's own internal tests, and the company itself admits there are no public extraction benchmarks to check it against. Until independent testers run Clairic on their own documents, the accuracy claim is a promise, not proof. The open-weight release helps here — anyone can now test it — but the verification has not happened yet.

What to watch next is adoption: whether developers pick Clairic up on Hugging Face and whether the promotional $0.10 pricing converts into paid usage after December 1. If the Workers AI superintelligence team keeps shipping, Cloudflare starts to look less like a networking company with an AI side project and more like a model provider in its own right. Reader poll: Bullish or bearish on Cloudflare after the Clairic launch? Vote: Bullish, Bearish, Mixed.

Are you bullish or bearish on this story?

By The Numbers

27BParameters in Clairic
9BParameters in Clairic-flash
207 to 0Image-reading errors on Cloudflare's own internal evals
$0.10Promotional price per million input tokens until Dec 1
$300.91NET close Wed, up 5.66%

FAQ

What does open-weight mean?

Open-weight means the model's trained parameters — its learned knowledge — are published for anyone to download and run. Unlike a closed model you can only use through a company's service, an open-weight model can be hosted, studied and built on by others, under the terms of its licence.

What is a token?

A token is a small chunk of text — roughly three-quarters of a word — that AI models read and write. Clairic charges only for input tokens, so $0.24 per million input tokens is the cost of feeding it about 750,000 words of documents.

What is document extraction?

Document extraction is pulling structured information out of messy files — text, tables, dates and figures from PDFs, scans and images — so software can use it. It is the unglamorous work behind invoices, contracts and forms, and getting it wrong means bad data downstream.

👁

Source: Cloudflare

Get the daily briefing. AI news with a markets lens — tickers, bullish/bearish reads, zero hype. Subscribe free on Substack

More on $NET

All $NET coverage →

More from AI Agent Update