
Cloudflare Releases Updated Open-Weight Clairic and Clairic-flash Models
Cloudflare introduced an updated Clairic extraction model and Clairic-flash, a 9-billion-parameter variant, with open weights on Hugging Face and hosting on Workers AI for developers.
Cloudflare has introduced an updated version of its Clairic extraction model and a new Clairic-flash model. Both models are open-weight, hosted on Workers AI, and available on Hugging Face. The release comes from the Workers AI superintelligence team, which is also behind Clef.
Redesigned for extraction
Clairic was redesigned after its preview release based on customer feedback. For ordinary text and number extraction fields, the model now uses parallel slot autoregressive generation, where each field uses tokens it has already written to choose its next token and stops at its own end marker. All fields advance together in the serving batch, keeping latency low while improving accuracy. Special heads handle estimates, forecasts, durations, calculated dates, and explicit arithmetic. The typed API interface still provides confidence scores, character spans, and probabilities.
Images, PDFs, and speed
The updated Clairic family can read PDFs as well as images, with support for up to four images or PDF pages in total. Line items can be returned as rows that keep each description, quantity, and price together. Image reading accuracy improved from 207 errors to 0 on internal benchmarks. In latency tests, complete HTTP responses were under one second, with Clairic-flash showing lower p50 and p95 times than Clairic across listed eval sets.
Pricing and availability
Clairic is available at @cf/cloudflare/clairic and Clairic-flash at @cf/cloudflare/clairic-flash. The 27-billion-parameter Clairic model is priced at $0.24 per million input tokens, while the 9-billion-parameter Clairic-flash model is priced at $0.10 per million input tokens. A promotional discount keeps both models at $0.10 per million input tokens until December 1, 2026. Cloudflare charges only for input tokens and absorbs output token generation costs. Existing applications using the preview endpoint automatically receive the updated model.
Sources: The Cloudflare Blog
SiTech — AI-powered web development
We build fast, modern websites and bring AI into real business workflows. Have a project or a question? We'd love to help.