Cloudflare launches Clairic for faster, cheaper data extraction
The Workers AI model extracts, computes, estimates and forecasts typed values without generating tokens.
Cloudflare has released Clairic in preview on Workers AI, introducing what it calls a “System One” extraction model. Unlike a general-purpose LLM, Clairic turns unstructured text or images into typed values, calculations, estimates and forecasts in one forward pass.
The model does not generate output tokens. It can return names, dates and lists with source spans, compute values not written explicitly, and provide estimates or forecasts with an 80% uncertainty interval. Cloudflare says requests typically finish in under a second and fields are processed in parallel.
Clairic is intended to sit before a decision model and an LLM in agent workflows. Extraction happens first, another model chooses an action, and an LLM writes the response. That reduces raw input seen by the LLM and makes intermediate results easier to audit.
In Cloudflare’s customer-claim test, a Clairic–Clef–GLM-5.3 pipeline finished in 19.3 seconds versus 61.7 seconds for an all-GLM-5.3 pipeline, with a calculated 49% lower cost. These are Cloudflare’s own figures. The endpoint is an early checkpoint: reading is strong, while multi-step arithmetic needs improvement. Open weights are planned.
