What changed
Cloudflare has announced the release of two new AI models: an updated version of Clairic and a lighter, faster variant called Clairic-flash.
The updated Clairic focuses on improved extraction from complex documents, including images and PDFs, with better handling of tables, lists, and dates. Clairic-flash is a 9-billion-parameter model that delivers strong extraction and prediction performance while being more compact and efficient.
Both models are hosted on Workers AI and their full open weights are available on Hugging Face. The updated Clairic uses a redesigned architecture with parallel slot autoregressive generation, which improves accuracy for text and image extraction without significantly increasing latency. The model also includes special heads for computation, estimation, and extraction use cases.
What this means in practice
The new models are intended for users who need to extract data from unstructured sources such as receipts, support tickets, and multi-page PDFs.
The improved architecture allows for faster and more accurate extraction, with results that include confidence scores, character spans, and probabilities.
The models are also designed to be cost-effective, with promotional pricing available until December 1, 2026, where both Clairic and Clairic-flash will cost $0.10 per M input tokens.
The models are available at the following endpoints: /cloudflare/clairic for the 27B model and /cloudflare/clairic-flash for the 9B model. Developers can access the full API schemas from the developer documentation, which remain unchanged from the preview release.
Limitations and availability
The supplied announcement does not specify public extraction benchmarks, though internal evaluations show improved performance across various use cases. The models are available on Workers AI and Hugging Face, with open weights for both Clairic and Clairic-flash.
The team behind these models, the Workers AI superintelligence team, is continuing to develop frontier AI research for Cloudflare’s customers and the open-source community.
