Focused on what LLMs do inefficiently
Cloudflare announced a new class of "System One extraction models." The first, Clairic, is available in preview on the Workers AI platform as @cf/cloudflare/clairic. The company plans to release the model's open weights soon.
Why a separate model class?
According to Cloudflare, many structured-output calls sent to frontier LLMs repeat the same shapes: What's the invoice total? When was this delivered? How many days late is it? These answers can't be listed in advance. The company built extraction models to fill that gap.
What it does
Clairic extracts, computes, estimates and forecasts values from unstructured inputs. Outputs come in four forms:
- Extracted values: Text copied from the input, such as a name or order ID, with character offsets and a confidence score.
- Computed values: Numbers, dates, times, durations and lists that follow their type rules, with a confidence score.
- Estimated values: Values the input doesn't state but gives evidence for, returned as a distribution within declared bounds with a median and an 80% interval.
- Forecasted values: Future values of a series, one estimate per step ahead, each with its own 80% interval.
Speed and cost
Cloudflare says Clairic answers in a single pass without reasoning or generating tokens. It charges only for input and typically responds in under a second, while an LLM reasoning through the same step takes several seconds. Because it generates no tokens, it can't produce a mistyped answer or hallucinate the way LLMs can, though it can still make mistakes.
Where it fits
Cloudflare says the idea was inspired by TypeSafe's System One models announced a month ago. Unlike decision models (Jev and Clef), Clairic returns values directly. Cloudflare positions Clairic for numbers, Clef for choosing among fixed options, and LLMs for open-ended text.
