**[Cloudflare Uses Three AI Models in Tandem for Claims Processing: 3.2x Faster, 49% Cost Reduction]** Beating AI Newsflash: Cloudflare has released a preview version of its data extraction model, Clairic. By providing it with an invoice or a customer service email, it can identify order numbers, dates, total amounts, estimate repair costs based on existing information, and even predict customer traffic. Unlike traditional LLMs that generate answers token by token, Clairic directly computes and returns results of specified types, along with confidence levels. Fields for which no basis can be found will return null. Clairic follows the System One model approach that gained traction after TypeSafe introduced Jev. On October 1, Cloudflare launched its decision-making model, Clef, which selects from preset options, while Clairic handles extraction, computation, and prediction of specific values. Neither model generates answers word by word. Clairic charges only $0.10 per million input tokens, with no fees for output tokens. Cloudflare demonstrated its use in processing claims for damaged goods: Clairic first extracts information and estimates repair costs, Clef decides whether to refund or repair, and GLM-5.3 finally drafts a response to the customer. The three models working together took 19.3 seconds, compared to 61.7 seconds if the task were handled solely by GLM-5.3. On a scale of one million claims, costs dropped from $11,740 to $6,032, a 49% reduction. Specifically, the information extraction phase saw processing time drop from 16.1 seconds to 0.46 seconds. However, while Clairic does not generate output tokens, if certain fields cannot be processed in a single computation, the system automatically splits the task into multiple parallel rounds, re-reading the full input for each round. For high-token inputs like images, this could increase actual invocation costs. The model is currently available for preview on Workers AI, though its weights have not yet been open-sourced. [Original Article Link]