VYPR
researchPublished Nov 18, 2026· Updated Nov 21, 2026· 1 source

Cloudflare Releases Open-Weight AI Models Clairic and Clairic-flash for Enhanced Data Extraction

Cloudflare's Workers AI Superintelligence team has launched updated open-weight AI models, Clairic and Clairic-flash, designed for advanced data extraction from complex documents and images.

Cloudflare has announced a significant advancement in its AI capabilities with the release of two open-weight models, Clairic and Clairic-flash, developed by its new Workers AI Superintelligence team. This initiative positions Cloudflare as a creator of frontier AI models for both its customers and the broader open-source community.

The updated Clairic model boasts substantial improvements over its preview version, particularly in its ability to extract data from complex documents, including images and PDFs. Key enhancements include more reliable handling of tables and dates, better grounding of extracted information in the source material, and a redesigned architecture that employs parallel slot autoregressive generation for improved accuracy in text and image extraction. Specialized heads have also been trained to manage estimations, forecasts, durations, calculated dates, and arithmetic operations.

Complementing the full Clairic model is Clairic-flash, a more compact and faster 9-billion-parameter version. Despite its smaller size, Clairic-flash delivers strong performance in extraction and prediction tasks, making it suitable for applications where speed and efficiency are paramount. Both models retain useful features from the typed API interface, such as confidence scores, character spans, and probabilities, ensuring the extracted data is actionable.

Cloudflare has conducted internal evaluations comparing its new models against other extraction solutions like Liquid AI's D1.2 and Scaledown's Extract. The results, while self-reported and directional, indicate that Clairic models excel across various extraction use cases, validating Cloudflare's hypothesis that data extraction remains a significant challenge for LLM users due to cost, latency, and accuracy issues.

Latency tests demonstrate that both Clairic and Clairic-flash can deliver complete HTTP responses in under one second, a notable speed advantage for similar workloads. This performance is achieved even with the redesigned architecture, ensuring rapid data retrieval for users.

The enhanced capabilities extend to image and PDF processing. Clairic can now process PDFs, allowing users to extract information like supplier names, line items, and totals from multiple pages in a single request. Image reading accuracy has also seen a dramatic improvement, with zero errors on internal benchmarks for text and images in the new version.

To encourage adoption, Cloudflare is offering promotional pricing until December 1, 2026. During this period, both the 27B Clairic model (priced at $0.24 per M input tokens) and the 9B Clairic-flash model (priced at $0.10 per M input tokens) will be available at the lower rate of $0.10 per M input tokens. This promotion allows users to experience the higher quality Clairic model at the price of the faster Clairic-flash variant.

Both Clairic and Clairic-flash are available on Cloudflare's Workers AI platform and as open weights on Hugging Face, providing developers with flexible options for integration. The Workers AI Superintelligence team continues to push the boundaries of AI research, with further innovations expected.

Synthesized by Vypr AI