Cloudflare has released a preview of Clairic, an extraction model that pulls typed values from messy text and images in a single pass. The model does not generate tokens, which sets it apart from typical LLM pipelines that produce natural language output.
In a test on a damaged-product support claim, a pipeline built around Clairic finished in 19.3 seconds. Cloudflare's announcement frames this as a way to avoid overpaying an LLM to read invoices and similar documents.
Because Clairic skips token generation, it avoids the computational cost of producing text. That could make it a practical option for document-heavy tasks where the goal is to extract specific fields rather than generate responses.