# Cloudflare previews Clairic, an open-weights extraction model for structured data

> **Key takeaway:** Cloudflare announced Clairic, a fast single-pass extraction model that pulls typed values from unstructured input without generating tokens, with open-weights release planned.

**Published:** 2026-10-19T17:48:26+00:00 
**Source:** Cloudflare Technical Blog 
**Category:** software-engineering 
**Canonical URL:** https://fosswire.org/news/cloudflare-previews-clairic-an-open-weights-extraction-model.html 

## Summary
Cloudflare has previewed Clairic, a new 'System One' extraction model hosted on Workers AI. Unlike LLMs, it answers in a single forward pass without reasoning or token generation, charging only for input tokens. It extracts, computes, estimates, and forecasts typed values from unstructured text and images. Cloudflare intends to release the model weights openly soon.

## How Clairic differs from LLMs
- **Single-pass architecture**: No output tokens or reasoning; processes input in one parallel forward pass, typically under one second
- **Typed outputs**: Returns text spans with character offsets, numbers, dates, times, durations, lists, estimates with 80% intervals, and forecasts
- **No hallucination in the LLM sense**: Cannot mistype answers since it doesn't generate tokens; outputs are values of declared types or null
- **Per-field confidence**: Every answer includes `p_null` (probability value isn't in input) and `confidence`; estimates include median and p10-p90 interval
- **Cost model**: Charged only for input tokens, no output token costs
- **Parallel field processing**: Unlimited fields split into parallel passes that each see full input, with merged answers

## Why it matters
If Cloudflare follows through with open-weights release, Clairic could provide a self-hostable alternative to LLM-based extraction pipelines, reducing latency and cost for document processing and forecasting workflows.