# Introducing Clairic: A New System One Extraction Model Cloudflare launches Clairic, a specialized model for high-speed, typed data extraction and forecasting. - **Source:** [Cloudflare Blog](https://blog.cloudflare.com) - **Published:** 2026-10-19 - **Topic:** [Data & AI](https://brifing.si/c/data-ai) - **Original article:** - **On Brifing:** https://brifing.si/s/36cd9585-bf1c-489f-95c9-0b99f0b2c57d ## New System One Paradigm Cloudflare has introduced Clairic, a System One-style extraction model designed for speed and precision. Unlike traditional LLMs that rely on deliberate reasoning, Clairic operates in a single parallel pass. This architecture allows it to extract, compute, and forecast values from unstructured inputs significantly faster and more cost-effectively than frontier generative models. ## Beyond Generative LLMs While LLMs excel at creative writing and complex reasoning, they are often inefficient for structured data extraction. Clairic addresses this by avoiding token generation entirely, which eliminates common hallucination risks. It provides typed outputs such as numbers, dates, and lists, ensuring that every result adheres to a strictly defined schema. ## Core Functional Capabilities Clairic handles three primary output types: extracted text spans, computed values, and estimated distributions. It returns character offsets for extracted data, allowing developers to trace answers back to the source. Computed values, such as sums or durations, are derived directly from the input, ensuring accuracy without requiring pre-sanitization of the raw data. ## Forecasting and Estimation The model provides estimated values with an 80% confidence interval, offering a distribution rather than a single, potentially misleading point. For series data, it can forecast future values step-by-step. This probabilistic approach allows downstream systems to assess the model's certainty, enabling automated workflows to trigger manual reviews when confidence scores are low. ## Single Pass Efficiency Clairic processes requests in a single forward pass, meaning there are no output token charges. Whether a user requests three fields or thirty, the computational cost remains consistent. This design makes it highly scalable for high-volume document processing tasks where latency and cost per request are critical operational constraints for developers. ## Handling Unstructured Inputs The model accepts text and up to four optional images as input states. By mapping fields with plain-language instructions, users can extract specific data points from messy sources like receipts or support emails. If a requested field is missing, the model returns a null value along with a probability score for that absence. ## Practical Extraction Example In a customer support scenario, Clairic can parse an email to identify order IDs, delivery dates, and damage descriptions. It can even perform arithmetic, such as adding tax to a base price, to return a total amount. The model provides confidence scores for each field, helping developers validate the extracted information automatically. ## Visual Data Processing Clairic extends its extraction capabilities to image inputs, such as photos of receipts. Despite potential issues like poor lighting or skewed angles, the model successfully identifies quantities, unit prices, and subtotals. It maintains its typed output structure, correctly identifying when certain fields, such as discounts, are absent from the provided image. ## Comparison with Decision Models Clairic complements existing System One decision models like Jev and Clef. While decision models output probabilities for specific choices, Clairic focuses on extracting and computing values. Together, these tools form a pipeline where extraction models prepare data, decision models evaluate options, and LLMs handle final communication tasks for complex workflows. ## Confidence and Reliability Every scalar answer from Clairic includes a p_null probability, indicating the likelihood that the value is missing from the input. This feature is vital for building robust pipelines that require high data integrity. By providing confidence metrics, the model allows developers to implement conditional logic based on the certainty of the extraction results. ## Integration and Availability Clairic is currently available in preview on Workers AI as @cf/cloudflare/clairic. Cloudflare intends to release the model with open weights in the near future. This accessibility allows developers to integrate high-speed extraction directly into their edge applications, reducing reliance on expensive, slow, and non-deterministic LLM calls for routine data parsing. ## Strategic Industry Impact The emergence of System One models signals a shift toward specialized, task-specific AI architectures. By decoupling extraction from reasoning, developers can optimize for specific performance metrics. This modular approach reduces the technical debt associated with complex prompt engineering and parsing layers, ultimately leading to more reliable and predictable automated business processes. ## Sources covering this story - https://blog.cloudflare.com/clairic-extraction-model/ — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._