Cloudflare released the weights of its Clairic 27B and Clairic-flash 9B document extraction models under Apache 2.0, the first public output from its Workers AI Superintelligence team.
Cloudflare released the full model weights for its Clairic document extraction AI this week, dropping two models on Hugging Face under the permissive Apache 2.0 license: Clairic 27B and the new Clairic-flash 9B. The release is the first public output from Cloudflare's rebranded Workers AI Superintelligence team and signals the company's intent to become a serious player in open-source frontier AI — not just a CDN and infrastructure provider.
Document extraction is the process of pulling structured data — names, dates, line items, totals — from unstructured documents like PDFs, scanned receipts, and invoices. It is one of the most commercially valuable but unglamorous AI tasks, underpinning accounts payable automation, expense management, and compliance workflows across industries.
What Cloudflare Released
According to Cloudflare's announcement, two models are now publicly available:
- Clairic 27B — the full-precision model, updated with improved accuracy on complex multi-page documents
- Clairic-flash 9B — a new, smaller model optimized for speed and lower inference cost, targeting high-throughput pipelines
Both are released under Apache 2.0, which means developers and companies can use, modify, and deploy them commercially without royalty or restriction. The weights are hosted on Hugging Face and available as of November 19, 2026.
Get this in your inbox.
Daily AI intelligence. Free. No spam.
The models were developed internally by Cloudflare's Workers AI Superintelligence team, a unit the company formally named earlier this year as it shifted from purely hosting third-party models to building its own.
Why This Matters
Cloudflare has 20 million internet properties running on its network. Embedding capable AI directly into its Workers AI platform — where developers already deploy serverless code — reduces the friction between "I have documents" and "I have structured data" to essentially zero. By open-sourcing the weights, Cloudflare also creates a feedback loop: developers who run Clairic locally will likely migrate production workloads to Workers AI for scale, latency, and pricing.
For the broader AI ecosystem, the release adds competitive pressure to closed-source document AI products from companies like Google (Document AI), AWS (Textract), and several specialized vendors. Apache 2.0 licensing is notably more permissive than the "research only" or "non-commercial" clauses that limit some open-weight releases — Clairic can go straight into production.
The Clairic-flash 9B model is particularly notable. At 9 billion parameters, it runs efficiently on a single GPU and is small enough to deploy on-premise for companies with strict data residency requirements — a category that includes most banks, insurers, healthcare systems, and government agencies.
What to Watch
Cloudflare has not announced whether Clairic will be integrated into its R2 object storage or Cloudflare AI Gateway products, but the architectural fit is obvious. Watch for a Workers AI integration announcement that lets developers call Clairic directly on documents stored in R2 with a single API call. If that ships, it will make Cloudflare a one-stop shop for document intelligence at scale — and further blur the line between infrastructure provider and AI company.
Did this help you understand AI better?
Your feedback helps us write more useful content.
Get tomorrow's AI briefing
Join readers who start their day with NexChron. Free, daily, no spam.