Cloudflare launches Clairic, a new System One extraction model for Workers AI
Cloudflare introduces Clairic, a fast, low-cost extraction model that extracts and forecasts data from unstructured inputs with higher accuracy than traditional LLMs.
Key points
- Clairic is the first System One extraction model, distinct from decision models like Jev.
- It processes inputs in under a second without generating tokens or reasoning chains.
- The model returns typed values, computed numbers, estimates, and forecasts with confidence scores.
Cloudflare announced a preview of Clairic, a new System One extraction model, on its blog. The model is available hosted on Workers AI and is designed to extract, compute, estimate, and forecast data from unstructured inputs. Cloudflare stated it intends to release the model's weights as open-source in the near future.
Clairic operates significantly faster and at a lower cost than traditional large language models. According to Cloudflare, it achieves higher accuracy for specific tasks by avoiding the reasoning steps required by LLMs. The model charges only for input tokens and typically responds in under a second, whereas an LLM reasoning through the same step takes several seconds. This speed and efficiency address inefficiencies in current workflows that rely on frontier LLMs for structured-output calls.
Context and Background
The concept of System One models was introduced by TypeSafe roughly a month ago, gaining industry attention. Based on Daniel Kahneman’s framework in Thinking, Fast and Slow, System One models are defined as fast and precise, contrasting with System Two models capable of deliberate reasoning. TypeSafe previously released Jev, a decision model purpose-fit for specific use cases, which prompted developers to consider non-LLM alternatives for their workflows.
Cloudflare identified tasks where LLMs are inefficient, such as calculating invoice totals, delivery dates, or repair costs. These tasks require values that cannot be listed in advance for probability-based decision models. Clairic represents a new class of System One extraction models that answer with direct values extracted from input, computed from it, or predicted beyond it, rather than selecting from fixed options.
Who Is Affected
Developers using frontier LLMs for structured data extraction may benefit from this new model class. Clairic handles outputs that are numbers, typed values like names and dates, or lists. It returns extracted text spans with character offsets and confidence scores. For computed values, it adheres to type rules, such as two decimals for currency, and provides confidence scores.
The model also provides estimated values based on evidence in the input, returning distributions with medians and 80% intervals. Forecasted values offer future estimates for series data, with each step ahead including its own 80% interval. Because Clairic generates no tokens, it cannot produce mistyped answers or hallucinate in the manner of generative LLMs, though it can still make mistakes.
What should you do?
Security and technology professionals should monitor the availability of open-weights for Clairic once Cloudflare releases them. Teams currently using LLMs for data extraction should evaluate whether their use cases involve fixed-structure outputs that could benefit from faster, token-free processing. Users should test the model's confidence scores to determine if the output reliability meets their specific application requirements before integration.
Background: Large language models
Large language models are a specific type of neural network trained on text. Generative AI is the broader umbrella term for any system that creates new content, including images and code. Not all generative AI uses large language models, but most modern text generators do.
What to do and how to stay safe: Cloudflare
- Review current LLM usage for simple extraction tasks to identify potential cost and latency savings.
- Monitor Cloudflare’s blog for the open-weights release schedule to assess self-hosting options.
- Evaluate the confidence scores provided by extraction models to ensure output reliability for critical data.
- Test integration with Workers AI to measure performance differences against existing generative models.
Step-by-step guide: Mobile Device Security: How to Protect Data Beyond Passwords
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
What is the difference between Clairic and traditional LLMs?
Clairic is a System One extraction model that answers in a single pass without reasoning or generating tokens, offering faster latency and lower cost than traditional LLMs.
Can Clairic hallucinate like a large language model?
Because Clairic generates no tokens, it cannot produce mistyped answers or hallucinate in the same way as an LLM, though it can still make mistakes.
When will Clairic be available as open weights?
Cloudflare intends to release the model open-weights soon, but no specific date has been provided in the announcement.
