Cloudflare announced on October 19, 2026 the preview of Clairic, the first model in a new class called System One extraction models. Available hosted on Workers AI as @cf/cloudflare/clairic, the model extracts, computes, estimates and forecasts values from unstructured inputs with higher accuracy, lower latency and lower cost than traditional LLMs. The company intends to release open weights soon.
What happened?
Inspired by the System One models announced by TypeSafe (such as Jev), Cloudflare developed Clairic for tasks where LLMs are inefficient: extracting typed values from texts or images. Unlike decision models (such as Clef), it returns concrete values — text spans, numbers, dates, lists, distributions — in a single forward pass, without generating output tokens.
This means there is no charge for output tokens, responses typically under one second, and no possibility of producing malformed JSON or hallucinations in the same way as LLMs. Each answer comes with a confidence score and null probability.
Why it matters
Extracting structured data from documents, emails, receipts or images is one of the most common workloads in agent and automation pipelines. LLMs do this with irregular success, requiring validation, retries and high cost. Clairic offers always-valid output for the requested type (or null), with traceable spans and image support. In an example from the official blog, it sums price + tax from an email and calculates days since delivery automatically.
Announced price: $0.10 per million input tokens. 65,536-token context. Can be used in pipelines with decision models (Clef) and LLMs for more complex actions.
Sources
- Official announcement on Cloudflare blog
- Clairic model documentation
- Original post on Cloudflare's X
Transparency: This content was created, edited or reviewed with the assistance of artificial intelligence. Information was cross-checked with public posts on X and sources available on the internet. Consult the original sources for the full context.
By GeekikiBot