Clairic: Wait on Cloudflare's Extraction Model
Cloudflare's Clairic fills document fields in one pass and looks cheaper than an LLM. The evidence is a vendor test with no accuracy numbers.
Wait. Clairic, Cloudflare's preview extraction model, looks promising for field extraction, but the only evidence is Cloudflare's own test: a three-step pipeline finished in 19.3 seconds versus 61.7 for GLM-5.3 alone, at roughly half the cost. No accuracy benchmark or open weights exist yet, so pilot it, don't migrate.
The Weights Desk · 3 min read- Bottom line: Wait. The one deciding number is Cloudflare's own 19.3 seconds versus 61.7 seconds for GLM-5.3 on a three-step pipeline.
- Clairic returns every requested field in a single pass, with a confidence score per value and null for missing data, rather than generating tokens.
- Neither the coverage nor Cloudflare's announcement, as summarized, includes an independent accuracy benchmark, so extraction quality on your documents is untested.
- Cloudflare says Clairic cannot yet reliably run longer calculation chains, so arithmetic-heavy logic should stay in your own code.
- Clairic is in preview on Workers AI. Cloudflare says open weights will follow but has given no date.
Wait. Cloudflare has put Clairic, an extraction model that fills named document fields in a single pass, into preview on Workers AI. The company's headline claim is that it beats a large language model on speed and cost. The number behind that claim comes from Cloudflare's own three-step test, and the coverage we read reports no accuracy benchmark. That is a promising lead, not yet a reason to rewire a production pipeline.
What Clairic does differently
Clairic does not write text token by token. Per The Decoder's report of Cloudflare's announcement, you name the fields you want, such as an order number or an invoice total, and the model answers all of them in one pass. Each value comes back in the requested format with a confidence score, and a missing value returns null. It can also handle simple calculations, estimates and forecasts, which makes it a component rather than a general assistant.
The one number: 19.3 seconds versus 61.7
The deciding figure is a pipeline comparison Cloudflare ran itself. In its example, a complaint email goes through three steps: Clairic extracts the order number, amount and damage and estimates repair cost, Clef applies company rules to pick a refund or repair, and GLM-5.3 drafts the customer reply. That chain took 19.3 seconds. GLM-5.3 alone took 61.7 seconds for the same three steps and cost almost twice as much.
What the test does not show
The comparison measures latency and cost, not correctness. The Decoder says the article gives no accuracy benchmark, parameter count or specific pricing. A faster model that misreads an invoice total is not cheaper once a human has to catch the error. Only one vendor-run scenario has been reported, so nobody outside Cloudflare has yet reproduced the result on their own documents.
Limits and availability
Cloudflare says Clairic cannot yet reliably handle longer calculation chains, and advises developers to run those in their own code. The model is a preview on Workers AI, and open weights are promised for later, with no date. That means you cannot yet run it on your own hardware, audit it, or fine-tune it, which matters for teams with data-residency or red-team requirements.
The bottom line: Wait, and pilot
The bottom line: Wait before migrating, but a small pilot is cheap. Take a labelled sample of your own documents, run Clairic and your current LLM side by side, and score field-level accuracy, null handling and confidence calibration. Also feed it hostile documents, since extraction models read untrusted text. Move to Use it only if accuracy holds and the speed and cost gap survives your workload. Until then, treat Cloudflare's 19.3 versus 61.7 seconds as a hypothesis.
- What is Clairic?
- Clairic is a Cloudflare extraction model in preview on Workers AI. You name the fields you want, and it returns typed values with confidence scores in a single pass instead of writing text token by token.
- Is Clairic really faster and cheaper than an LLM?
- Cloudflare reports that its three-step Clairic pipeline took 19.3 seconds against 61.7 seconds for GLM-5.3 alone, at about half the cost. This is a single vendor test, not an independent benchmark.
- What is the bottom line on adopting Clairic?
- Wait. Pilot it on a labelled sample of your own documents and measure field-level accuracy first. No accuracy figures, parameter count or open weights have been published yet.