Cloudflare Introduces Clairic System One Extraction Model
Cloudflare has announced a preview of Clairic, marking the introduction of a new class of System One extraction models hosted on Workers AI.
Radwan AbdulhadiEditor and technology writer
Introduction of System One Extraction Models
Cloudflare has announced a preview of a new class of System One-style models focused on data extraction. The company's first model in this category, named Clairic, is currently available hosted on Workers AI as @cf/cloudflare/clairic, with open-weights expected to be released soon. Detailed information regarding the new architecture and its capabilities can be found through the Cloudflare Blog.
The release builds upon recent industry shifts toward fast, precise architectures. When System One models were announced by TypeSafe a month ago, the concept captured significant industry attention. As referenced by TypeSafe in their blog and defined in Daniel Kahneman’s book Thinking, Fast and Slow, System One models prioritize speed and precision, offering a direct contrast to System Two models like large language models that are capable of deliberate reasoning.
Architecture and Operational Differences from LLMs
Unlike traditional large language models that generate responses token by token through extensive reasoning steps, Clairic processes input tokens in a single forward pass. Because it generates no output tokens, the model avoids traditional text generation overhead and eliminates token-based generation errors or standard hallucinations, though it can still make operational mistakes. Users are charged solely for input processing, and responses are typically delivered in under a second.
Cloudflare designed the model to handle tasks where traditional frontier LLMs can be inefficient. While structured-output requests frequently hit LLMs for recurring queries such as invoice totals, delivery dates, or repair costs, these tasks often lack pre-listed answers required by standard decision models. Clairic addresses this gap by functioning as a dedicated extraction engine that operates swiftly and cost-effectively.
Capabilities and Output Types
While previous System One models primarily functioned as decision engines—such as Clef and Jev, which evaluate input states against lists of questions to return probabilities—Clairic introduces a distinct operational paradigm. It retrieves exact text excerpts, computes numerical metrics, generates estimations within bounded ranges, and forecasts future data points based on unstructured inputs.
The model's output architecture supports multiple specialized data types. Extracted values return specific text spans complete with character offsets and confidence scores. Computed values cover numbers, dates, times, durations, and lists that adhere to strict type rules. Estimated values provide distributions within user-defined bounds, featuring medians and 80% intervals, while forecasted values project future series trends step by step.
Integration and Workflow Applications
System One extraction models, decision models, and System Two large language models are designed to complement one another within broader data workflows. Extraction models handle the raw input conversion into typed values, decision models translate those values into actionable choices, and LLMs ultimately convert decisions into conversational words.
Developers seeking to evaluate the practical implementation of these tools can explore the live demo you can try here, which showcases how extraction pipelines handle unstructured inputs. Additional technical context is also available through a new docs explainer detailing how extraction models integrate into modern development stacks.
Sources
- Cloudflare BlogIntroducing Clairic, a new System One class of extraction models
Continue chronologically
Related entity coverage
Related stories
Cloudflare Adds Open European Models and Agnostic AI Defenses
Cloudflare expands its Workers AI platform with Europe's EuroLLM and Switzerland's Apertus while launching model-agnostic cyber defense workshops.
Cloudflare AI Search Launches Into General Availability
Cloudflare's AI Search reaches general availability, introducing native image embeddings, PDF OCR, larger file limits, and new billing terms.
Cloudflare Releases Open-Source Clef Decision Models
Cloudflare introduces Clef and Clef-flash, open-source decision models hosted on Workers AI alongside a new reinforcement learning platform.
Cloudflare ML uncovers evasive storefront attacks
Cloudflare says its Page Shield ML detected eight malicious JavaScript payloads that other scanning services failed to flag in live storefront traffic.
