Cloudflare Debuts Clairic AI Model to Cut Costs, Speed Data Processing

Cloudflare has unveiled Clairic, a specialized AI model created to extract information from documents and images faster and at lower cost than traditional LLMs. The new model is geared for enterprise workloads that require processing large volumes of data, ranging from financial documents to automated support systems.

Cloudflare describes Clairic as a System One model, an emerging category of specialized AI models built to perform specific tasks without the multi-step reasoning used by some LLMs.

In a company test involving customer service automation, Cloudflare claims a workflow combining Clairic with other AI models reduced processing time by 69% compared with an LLM-only approach. These company-reported results have not been independently verified.

The cost difference was also substantial, the company said. Cloudflare reported processing costs of $6,032 per million customer claims, compared with $11,740 using the GLM-5.3 language model alone.

This technology could create a shift toward using smaller, specialized models for routine tasks while reserving larger models for work requiring more complex language capabilities.

Under the Hood

Unlike conventional LLMs, which generate responses sequentially, Clairic processes requested information in a single operation. It can identify names, dates, prices and other data points, perform basic calculations and estimate numerical values.

The model also supports images, enabling businesses to extract information from photographed receipts and other documents.

A key feature is its ability to indicate uncertainty. Clairic provides confidence scores for extracted values and probability ranges for estimates. When information cannot be established from the input, the model can return a null value rather than give an unsupported answer. This capability could help companies limit errors when automating data-intensive processes, though Cloudflare acknowledges that properly formatted outputs don’t guarantee accuracy.

The model supports a context window of 65,536 tokens, with pricing starting at $0.10 per million input tokens. Because Clairic does not generate output tokens, customers pay only for input processing. Clairic is available in preview through Cloudflare’s Workers AI platform.

Combining Specialized AI Models

Cloudflare is positioning Clairic as one component of a broader AI architecture rather than a replacement for LLMs.

In the company’s customer service demo, Clairic extracts transaction details and estimates repair expenses. Cloudflare’s Clef decision model then evaluates the information against business rules, while GLM-5.3 generates the customer response. The combined workflow required 19.3 seconds, compared with 61.7 seconds for GLM-5.3 alone.

Clef belongs to the same System One category as TypeSafe’s Jev model, introduced in September. Both focus on selecting among defined options, though Clairic handles data extraction and numerical predictions.

The approach allows enterprises to assign different stages of an AI workflow to models designed for those specific functions.

Early Results and Limitations

Cloudflare tested Clairic’s ability to forecast levels of daily passenger traffic at U.S. airport security checkpoints. Over a 21-day period, its predictions differed from actual TSA passenger screening counts by an average of 3.3%. Actual passenger counts fell within the model’s predicted range on 17 of those days. So the system isn’t perfect.

At this point the technology remains in an early stage. Cloudflare cautions that Clairic struggles with calculations involving multiple dependent steps and is not intended to replace dedicated forecasting systems. The company recommends handling complex arithmetic through conventional software and reviewing uncertain predictions before using them in automated decisions.

Cloudflare plans to improve the model’s calculation and forecasting capabilities and release its weights publicly, though no date has been announced.