Cloudflare Just Open-Weighted Its Document Extraction Models — Clairic 27B and a 9B Flash
Nothing here is sponsored — no affiliate links on this page. Every fact below comes from Cloudflare's announcement of November 18, 2026. Benchmarks, latency figures and pricing are Cloudflare's own claims, not independently verified — and there are no public extraction benchmarks yet. We have not tested Clairic or Clairic-flash hands-on — this is researched analysis, not a review.
If you have ever pointed a big language model at a pile of PDFs and watched it confidently invent a total that isn't on any page, you already know why this launch matters. Cloudflare announced on November 18 that its document extraction models — Clairic and the new Clairic-flash — are now open-weight, on Hugging Face under Apache 2.0, and hosted on Workers AI. This is the graduation of the preview Cloudflare shipped on October 19, which promised open weights later. Later is now.
🔍 How we put this together
We used Cloudflare's official November 18, 2026 announcement — "An updated Clairic and Clairic-flash are now open-weight" — plus the Unite.AI coverage of the release. Where a figure is Cloudflare's own claim — the benchmark scores, the latency numbers, the pricing, the capability list — we say so. Nothing here is independently verified yet, and we have not run Clairic or Clairic-flash ourselves.What actually changed since the preview
The preview model could read a photo of a receipt. The new Clairic reads PDFs too — supplier, line items and total in one request, even when they sit on different pages. You can send up to four images or PDF pages in a single request, and line items come back as rows that keep each description, quantity and price together, instead of a bag of disconnected values.
Three things changed under the hood:
- A redesigned architecture. For ordinary text and number fields, Clairic now uses parallel slot autoregressive generation instead of a single forward pass: each field uses the tokens it has already written to choose its next token, stops at its own end marker, and all fields advance together in the serving batch. The trade-off is a small latency cost for a real accuracy gain.
- Special heads for computation. The model got trained heads that handle estimates, forecasts, durations, calculated dates and explicit arithmetic — the stuff that turns "extract the numbers" into "give me the answer the numbers imply."
- Better grounding. Answers stay tied to the source material, with the typed API interface intact: confidence scores, character spans and probabilities still come back with every request, so you can check where a value came from instead of taking it on faith.
Cloudflare says image reading accuracy improved from 207 errors to zero on its internal benchmarks. That number deserves the usual caveat — see the honest limits below — but the direction is the right one: extraction that fails silently on a scanned invoice is worse than extraction that fails loudly.
Two models, two jobs
- Clairic — 27B parameters (base Qwen3.8-27B). Maximum extraction accuracy on hard documents. The model you reach for when the document is messy and the answer has to be right.
- Clairic-flash — 9B parameters (base Qwen3.5-9B). The lighter, faster sibling: strong extraction and prediction performance in a compact model, built for high-volume, cost-sensitive workloads.
On Cloudflare's own latency tests — complete HTTP response from a self-hosted server, all fields included — Clairic answers fresh text fields in 358ms at p50 (736ms at p95), while Clairic-flash does the same in 155ms (328ms at p95). For loan pricing workloads, flash sits at 188ms p50. Both stay under a second, which is the whole point: this is a model you call inside a request handler, not a batch job you run overnight.
The numbers — and what they actually mean
Cloudflare is unusually upfront that there are no public extraction benchmarks yet. The numbers below are its own internal evals, run against other emerging extraction models (Liquid AI's D1.2 pipelined with Apple Vision OCR, and ScaleDown's Extract). Read them as directional, not proof:
- Synthetic receipt fields, text and images: 576/576 for both new models, up from 369/576 for the preview model. The rival pipelines scored 397/576 (D1.2 + OCR) and 280/576 (ScaleDown).
- Fresh text fields, 96 requests: 335/336 for Clairic, 328/336 for flash, vs 295/336 for the preview.
- Tracked customer assertions, 139 request shapes: 1,263/1,263 for both new models, vs 1,033/1,263 for the preview.
- Mixed-input fields: 221/226 for Clairic, 213/226 for flash, vs 207/226 for the preview.
The pattern is consistent across Cloudflare's own test sets — the redesign helped everywhere, and flash is close behind the big model on most sets. What the pattern cannot tell you is how either model handles your documents: your layouts, your handwriting, your edge cases. That test is yours to run.
The price — read this part carefully
Both models cost $0.10 per million input tokens until December 1, 2026. From December 1, Clairic (27B) rises to $0.24 per million input tokens, while Clairic-flash stays at $0.10. And there is no output-token charge at all — the models do generate output tokens, but Cloudflare only bills on input and eats the output cost.
One migration note: during the preview, the 9B model lived at the @cf/cloudflare/clairic endpoint. Now the 9B moves to @cf/cloudflare/clairic-flash and the 27B takes over @cf/cloudflare/clairic. If your app is already hooked up to the preview endpoint, you automatically get the better model — Cloudflare is rolling the new 27B in place. If you want the cheaper 9B going forward, point at the flash endpoint explicitly before December 1, when the price gap appears.
How to actually use them
On Workers AI (hosted, fastest path): @cf/cloudflare/clairic for the 27B model, @cf/cloudflare/clairic-flash for the 9B. The API schemas are unchanged from the preview endpoint, so existing integrations should keep working. You define the fields you want — typed values, not free text — and the model returns them with confidence scores and character spans pointing back at the source.
Self-hosted (open weights): Both models are on Hugging Face as open weights under Apache 2.0, so you can run them locally, fine-tune them, or deploy them on your own infrastructure. That matters if your documents can't leave your network — medical records, legal files, financial statements — or if you want to tune extraction for one specific document type and stop paying per token entirely.
What to build with them first: receipt and invoice ingestion (supplier, line items, totals, dates, grounded in the document); support ticket triage (pull structured values like order numbers, dates and amounts out of free text); lease and loan document processing (Cloudflare's own latency tests used commercial leases and loan pricing as eval sets, which tells you where they expect the model to earn its keep); expense audits and traffic forecasting (the estimation and forecasting heads are built for exactly this).
Who this fits — and who should wait
Clairic fits teams paying frontier-model prices to read documents today — invoice pipelines, ticket triage, lease and loan processing — where typed outputs with confidence scores and source spans would replace a human checking every field by hand.
Wait, or pilot carefully, if: your documents are unusual enough that vendor evals mean little (handwriting-heavy, low-quality scans, rare layouts); you need a fully open reproduction including training data, not just open weights; or your volume only makes sense at the $0.10 promo price and you haven't budgeted for $0.24 from December.
The honest limits
- No public extraction benchmarks exist. Cloudflare says so itself. Every score above is self-reported, on Cloudflare's own evals, against competitors Cloudflare itself ran. Independent testing is what turns a launch post into a recommendation, and it hasn't happened yet.
- The evals are Cloudflare's home turf. The test sets (receipts, leases, loan pricing, expense audits) are the workloads Cloudflare cares about. Your documents may behave differently.
- Open weights, not open training data. Apache 2.0 on the weights is genuinely permissive, but this isn't a fully open reproduction of the model. Fine-tuning is yours to do; the training recipe isn't.
- The promo ends December 1. The 27B model at $0.10 is a trial price. Budget for $0.24 from December, or plan your high-volume workloads on flash.
What to do this week
If you pay a frontier model to read documents today, the move is a small, measured pilot — not a wholesale swap:
1. Try the hosted endpoint first — @cf/cloudflare/clairic on Workers AI, same schema as the preview, no setup. Feed it your ugliest real document, not a clean demo receipt.
2. Compare both sizes on your own data — run the same documents through Clairic and Clairic-flash and check where the 9B model's answers diverge. The $0.10 promo price makes the test nearly free until December 1.
3. Use the typed outputs — confidence scores and character spans are the whole point. Build a rule: low confidence or missing span means human review, not silent acceptance.
4. If your documents can't leave your network, pull the Hugging Face weights — Apache 2.0, self-host, and test locally.
5. Decide before December 1 — that's when the 27B price more than doubles to $0.24. If your volume justifies it, great; if not, move the high-volume work to flash at $0.10.
Sources: Cloudflare, "An updated Clairic and Clairic-flash are now open-weight," November 18, 2026; Unite.AI, "Cloudflare Releases Open-Weight Clairic Models for Data Extraction," November 2026. Benchmark scores, latency figures and pricing are Cloudflare's own claims; no public extraction benchmarks exist yet, and the models have not been independently verified. Full version with FAQ on this page.