Cloudflare previews Clairic, a model that pulls typed values out of text and images

Cloudflare announced a preview of Clairic on October 19, 2026, a new kind of model for Workers AI that extracts data instead of writing text. You hand it some input, such as an email, a receipt photo or a table, plus a list of named fields with types like number, date, list or estimate. It answers every field in one quick pass, with no reasoning step and no generated tokens.

Because nothing is written out word by word, each answer is always the type you asked for, or null when the value is missing. Text answers come with their exact position in the input, and each result carries a confidence score. Clairic can also do simple math, such as adding a price and tax, estimate values with an 80% range, and make basic time series forecasts. In Cloudflare's damaged order demo, a pipeline using Clairic, its Clef decision model and an LLM for the final reply was about 3.2 times faster and 49% cheaper than using an LLM for every step.

Pricing is $0.10 per million input tokens with no output charge. Cloudflare admits the early checkpoint struggles with long chains of arithmetic and is not a replacement for dedicated forecasting models. It plans to release open weights later.

Why it matters

Pulling fields out of documents is one of the most common jobs given to LLMs, and a faster, cheaper model that cannot return malformed output could simplify a lot of data pipelines.

Source: Cloudflare Blog