Skip to content

Cloudflare open-sources Clairic, adds a 9B Clairic-flash model

Clairic leaves preview as a 27B extraction model with open weights on Hugging Face, joined by a new 9B Clairic-flash, both discounted to $0.10 per million input tokens on Workers AI until December.

Source: Cloudflare

Cloudflare published open weights for Clairic, the extraction model it introduced in October, and added a smaller sibling called Clairic-flash. Both now have full weights on Hugging Face under an Apache 2.0 licence, so you can run them with Transformers, vLLM or SGLang instead of going through Workers AI at all.

Clairic is a 27-billion-parameter model built on Qwen3.8-27B and moves from preview to general availability. Clairic-flash is new, at 9 billion parameters on a Qwen3.5-9B base. Instead of generating text, both take a typed-field schema and return structured values with confidence scores, pulling data out of documents, images and PDFs. On Workers AI they normally run @cf/cloudflare/clairic and @cf/cloudflare/clairic-flash at $0.24 and $0.10 per million input tokens, but until 1 December both are priced at $0.10, with no charge for the generated output.

Why it matters: if you’re pulling fields out of invoices, receipts or forms, this is a cheaper and more predictable alternative to prompting a general-purpose LLM for JSON, and open weights mean you can self-host rather than route every document through someone else’s API.

The caveat: the $0.10 promotional price ends 1 December 2026, when Clairic reverts to $0.24 per million input tokens.

Share

Related reading

More from The Wire

All briefs

A Linux KVM bug let one VM read 127GB of host memory

A bug bounty researcher used a four-year-old stale pointer in nested virtualization to read host kernel memory from inside an ordinary cloud VM. DigitalOcean disclosed and patched it as CVE-2026-99312.

via DigitalOcean