# Cloudflare's Three-Model Workflow Cuts Claim Processing Time by 69% in Test By Simon Yoon Canonical URL: https://www.tokenpost.com/news/technology/35968 Published: 2026-10-20T21:02:25.000Z Updated: 2026-10-20T21:02:25.000Z Section: Technology > Projected costs were $6,032 per million claims, compared with $11,740 when GLM-5.3 handled every step. Cloudflare's new Clairic model helped a three-model AI workflow process a damaged-product claim in 19.3 seconds, 69% less time than a GLM-5.3-only setup, in a test with lower projected costs. The workflow was about 3.2 times faster than the version that used GLM-5.3 for every step, which took 61.7 seconds. Projected costs per million claims were $6,032 for the three-model setup and $11,740 for the GLM-5.3-only setup, a 49% reduction. Those cost figures are estimates scaled to one million claims, not the cost of processing that many claims in the test. Cloudflare divided the task among three models. Clairic read a customer's email, extracted order and damage details, and estimated the repair cost. Clef selected a refund or repair under a specified policy, and GLM-5.3 drafted the customer reply. The extraction step took Clairic 0.46 seconds, compared with 16.1 seconds for GLM-5.3. GLM-5.3 wrote the reply in both versions of the workflow, so the comparison changed which models handled extraction and the refund-or-repair decision. Clairic is available in preview on Cloudflare's Workers AI service. Users specify the values they want returned, such as dates or amounts. The model returns those values with confidence information without generating a response one token at a time. It can return a null value when the input does not establish an answer. Cloudflare charges $0.10 per million input tokens for Clairic and no output-token fee. A request may require multiple processing passes if its fields cannot be handled in one pass. Each pass reads the full input again, increasing token use and potentially raising the cost of inputs such as images. Cloudflare introduced [Clef, its model for selecting from predefined choices](), on Oct. 1. In the claims test, Clef made the policy decision after Clairic supplied the claim details. Cloudflare plans to release Clairic's model weights later. ## Links in this article - [Clef, its model for selecting from predefined choices](https://www.tokenpost.com/news/technology/26238)