Cloudflare releases open-weight Clairic models for extracting data from documents

Nov 18, 2026
A 9-billion-parameter Clairic-flash joins the updated Clairic on Workers AI and Hugging Face.

Cloudflare has released two open-weight extraction models, an updated Clairic and a lighter, faster Clairic-flash with 9 billion parameters, and says it has become a "neolab" building its own frontier AI models for customers and the open-source community.

Both models pull structured data from text, images and PDFs, such as values from receipts or support tickets, and are hosted on Cloudflare's Workers AI platform, with full weights on Hugging Face. Cloudflare says the new Clairic improves handling of tables, lists and dates and grounds answers more reliably in the source material.

After feedback on the preview release, the team redesigned the architecture: ordinary text and number fields now use parallel slot autoregressive generation, where each field builds on the tokens it has already written, and special heads handle estimates, forecasts, durations, calculated dates and arithmetic. The models still return confidence scores, character spans and probabilities with every answer.

In Cloudflare's own evaluations, Clairic answered 335 of 336 fresh text fields correctly, against 295 for the preview version. The work comes from what Cloudflare now calls its Workers AI superintelligence team, which is also behind its Clef model, according to Cloudflare.

We hope you enjoyed this article

Consider subscribing to one of our newsletters like Silicon Brief or Daily AI Brief.

Also, consider following us on social media:

Free newsletter

Silicon Brief

Weekly coverage of AI hardware developments including chips, GPUs, cloud platforms, and data center technology.

Whitepaper

Tensordyne Napier: What If One Rack Could Do the Work of Nine?

Tensordyne

This Tensordyne whitepaper presents Napier, an inference-focused AI processor and rack-scale system based on the company’s TDN Math logarithmic number system. It examines infrastructure requirements for large mixture-of-experts and agentic models, compares major inference architecture approaches, and details the TDN AIP processor, TDN72 pod, TDN Link fabric, and Napier Ultra configuration. The paper reports simulation-based performance, cost, and accuracy-validation results, including Tensordyne’s projected comparison of one Napier rack with a nine-rack Nvidia Rubin plus Groq deployment; the chip is reported as taped out and in fabrication.

Read more
Free, six days a week

Daily AI Brief: the AI news that matters, in your inbox.