Key points
- Clairic is the first System One extraction model, designed for speed and precision over reasoning.
- It extracts, computes, estimates, and forecasts values from unstructured inputs with high accuracy.
- The model charges only for input tokens and responds in under a second, unlike traditional LLMs.
The short version
- Cloudflare released Clairic, a preview of a new System One extraction model hosted on Workers AI.
- The model extracts structured data from unstructured text without generating tokens or reasoning steps.
- It offers faster latency and lower costs compared to frontier large language models for specific tasks.
The details
Cloudflare announced a preview of Clairic, a new class of System One extraction models available on Workers AI. According to the company's blog, Clairic is designed to extract, compute, estimate, and forecast values from unstructured inputs. It operates with significantly higher accuracy, faster latency, and lower cost than traditional large language models. The company intends to release the model's weights openly in the near future.
Clairic differs from decision models like Jev and Clef, which provide probabilities against listed questions. Instead, Clairic returns direct values extracted from input, computed from it, or predicted beyond it. It handles outputs such as numbers, names, dates, and lists. The model provides confidence scores for each output to help users evaluate reliability.
Because Clairic does not generate tokens, it avoids the hallucination risks associated with large language models. Outputs are strictly typed values or null if no answer is found. The model charges only for input tokens and typically responds in under a second. This contrasts with large language models that may take several seconds to reason through similar steps.
The context
The concept of System One models was introduced by TypeSafe a month ago, gaining industry attention. Defined by Daniel Kahneman’s book Thinking, Fast and Slow, System One models are fast and precise, contrasting with System Two models capable of deliberate reasoning. TypeSafe’s Jev model demonstrated success in niche domains where large language models struggled.
Cloudflare identified inefficiencies in using frontier large language models for repetitive structured-output tasks. Common queries include invoice totals, delivery dates, and repair costs. These tasks do not require probabilistic decision-making but need direct value extraction. Clairic addresses this gap by providing a purpose-fit model for extraction tasks.
What to watch
- Monitor for the open-weights release of the Clairic model by Cloudflare.
- Watch for performance comparisons between Clairic and traditional large language models in extraction tasks.
- Observe how other vendors respond to the System One model category and extraction capabilities.
What it means for organisations
Teams should assess whether current large language model usage for data extraction can be replaced by faster, cheaper alternatives. Evaluate if tasks like invoice processing or log parsing benefit from System One models. Discuss with engineering leads how to integrate new model classes into existing workflows. Review supplier contracts to understand cost structures for input-only models. Ensure data privacy policies align with using new extraction tools.
What to do and how to stay safe: Cloudflare
- Audit current large language model usage for repetitive extraction tasks like parsing invoices or logs.
- Test Clairic in a non-production environment to compare speed and accuracy against existing solutions.
- Review data handling policies to ensure compliance when processing unstructured inputs with new models.
- Monitor vendor announcements for open-weights releases to evaluate self-hosting options for sensitive data.
Step-by-step guide: Secure Home Routers for Small Business Networks
General security guidance from the TechSec Headlines newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
What is Clairic?
Clairic is a System One extraction model by Cloudflare that extracts, computes, estimates, and forecasts values from unstructured inputs.
How does Clairic differ from large language models?
Clairic is faster, cheaper, and does not generate tokens or reason, reducing hallucination risks compared to large language models.
Is Clairic open source?
Cloudflare intends to release the model's weights openly, but it is currently available as a preview on Workers AI.
