Cloudflare released open weights for two extraction models on November 18 and rebranded its internal AI research group as the Workers AI superintelligence team, positioning the infrastructure company as a model builder rather than solely a hosting provider1.
The two models — an updated Clairic and a new Clairic-flash — are hosted on Workers AI and available for download on Hugging Face Inc.. Clairic-flash is a 9-billion-parameter model designed for speed, while the larger Clairic targets higher-fidelity extraction from complex documents.
Document and image extraction
The updated Clairic adds PDF and image reading to its extraction capabilities, handling up to four images or PDF pages per request. Cloudflare said image reading accuracy improved from 207 errors to zero on its internal benchmarks. The model can return structured data — supplier names, line items, and totals — from multi-page documents in a single call.
Both models include confidence scores, character spans, and probabilities in their typed API interface, and Cloudflare said they can handle estimates, forecasts, durations, calculated dates, and explicit arithmetic. Cloudflare said both models respond in under one second on the latency tests described in its blog post.
Pricing and availability
Clairic is priced at $0.24 per million input tokens and Clairic-flash at $0.10 per million input tokens. Both models carry a promotional discount of $0.10 per million input tokens until December 1, 2026, when Clairic reverts to its standard rate. The models are accessible on Workers AI as @cf/cloudflare/clairic and @cf/cloudflare/clairic-flash.
The 'superintelligence' team
Cloudflare described itself as having become "a neolab" and re-introduced the Workers AI superintelligence team as the frontier AI research group behind CLEF, Clairic, and future projects. The blog post said the team is working to "push the AI frontier and build useful models for Cloudflare's customers and the open source community".
ANALYSIS Cloudflare's move from hosting third-party models on Workers AI to publishing its own open-weight extraction models marks a vertical expansion — the company now competes with the model vendors it serves. The sub-dollar-per-million-token pricing on Clairic-flash, paired with sub-second latency claims, targets high-volume document-processing pipelines where cost and speed outweigh generalist capability.