TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Cloudflare has made Clairic available in preview on Workers AI, presenting it as a model for extracting requested fields from documents in one pass. In a company-reported example, a workflow using Clairic took 19.3 seconds, compared with 61.7 seconds for GLM-5.3 alone, and cost almost half as much; independent comparisons and broader performance details were not provided.
Cloudflare has released Clairic in preview on its Workers AI platform, describing it as a model that extracts specified values from documents in a single pass rather than generating text token by token. The company says Clairic can make document-processing workflows faster and less costly than using a large language model (LLM) alone, though the performance comparison cited so far is a Cloudflare example, not an independent test.
Developers specify the fields they need—such as an order number or invoice total—and Clairic returns corresponding values in the requested format, along with a confidence score. According to The Decoder’s report on Cloudflare’s release, the model returns null when a value is missing. Cloudflare also says it can perform simple calculations, estimates and forecasts alongside extraction from unstructured text.
Cloudflare proposes using Clairic as one component in a sequence of AI systems, rather than as a replacement for every language model. In the company’s complaint-email example, Clairic extracts an order number, amount and description of damage, then estimates repair cost. A decision model called Clef applies company rules to select a refund or repair, and GLM-5.3 drafts the customer reply.
Cloudflare says that three-part workflow takes 19.3 seconds. It reports that GLM-5.3 handling all three steps by itself takes 61.7 seconds and costs almost twice as much. The source does not provide the underlying prices, test conditions or an independent replication, so those figures describe Cloudflare’s example rather than a general benchmark.
A Faster First Step in AI Workflows
Document extraction is a common early stage in workflows that process invoices, customer complaints and other business records. A system that returns only requested fields could reduce the amount of text a general-purpose model needs to generate, while giving later software a structured result to act on. That may make it easier for developers to combine data extraction, policy decisions and written responses as separate tasks.
The practical case for Clairic depends on whether the company’s reported speed and cost gains hold across different documents and workloads. If they do, organizations may be able to reserve larger models for tasks such as drafting, while using a more specialized model to collect information. Cloudflare’s stated limitation on longer calculations also means the model is not presented as a standalone replacement for business logic or general-purpose AI.
As an affiliate, we earn on qualifying purchases.
How Cloudflare Splits the Work
Cloudflare’s approach separates tasks that are often bundled into a single LLM request. Clairic is intended to extract and format values; Clef handles rule-based decisions; and a language model such as GLM-5.3 writes natural-language output. The example illustrates how a developer might pass information between models and software components in an agent chain.
The distinction is also in how Clairic responds. Rather than producing a sequence of text tokens, it is described as answering each requested field in a single pass. Cloudflare says open weights are planned to follow the preview, but the report does not give a release date or further details about availability.
“The Clairic workflow takes 19.3 seconds, while GLM-5.3 alone takes 61.7 seconds for all three steps and costs almost twice as much.”
— Cloudflare, as reported by The Decoder
As an affiliate, we earn on qualifying purchases.
Benchmark Scope and Calculation Limits
The reported timing and cost figures are Cloudflare’s claims for one illustrated workflow. The source material does not specify how many runs were measured, which hardware or pricing assumptions were used, or whether the comparison covered a range of document types. It also does not report an independent evaluation of extraction accuracy or confidence scores.
It remains unclear how Clairic performs on noisy scans, unusual layouts or documents with ambiguous values. Cloudflare acknowledges a current weakness in longer chains of calculations and recommends that developers run those calculations in their own code. Details such as the model’s size, open-weight licensing terms and the timing of the planned open-weight release are not given in the report.
structured data extraction from PDFs
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Preview Testing and Open Weights
Developers can try Clairic through its Workers AI preview, according to the release described by The Decoder. Cloudflare has said open weights are to follow, but no date is stated. The next useful evidence will include fuller product documentation and tests showing how extraction accuracy, speed and cost vary across real-world workloads.
For now, developers considering the model will need to check whether it handles their document formats and fields reliably, and keep complex calculations in application code. Cloudflare’s workflow example sets out a possible division of labor among models, but its results should not be treated as a universal comparison until broader testing is available.
business document analysis software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is Cloudflare’s Clairic?
Clairic is a model for extracting requested information from documents and unstructured text. Cloudflare says it returns values in the requested format and provides a confidence score.
Where can developers access Clairic?
Clairic is available in preview on Cloudflare Workers AI. The report says open weights are expected to follow, but does not provide a release date.
How much faster is Clairic, according to Cloudflare?
In Cloudflare’s example, a workflow using Clairic took 19.3 seconds, compared with 61.7 seconds for GLM-5.3 to handle all three steps alone. This is a company-reported example, not an independently verified benchmark.
Can Clairic replace an LLM or handle all calculations?
Cloudflare presents Clairic as one component in a larger workflow, with a language model handling tasks such as drafting a reply. The company says Clairic can do simple calculations but cannot yet reliably handle longer calculation chains; it recommends using developers’ own code for those.
Source: rss
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
