Cloudflare launches Clairic, a ‘System One’ model that extracts, computes and forecasts values in one forward pass without generating tokens
Confirmed
The takeaway
In Cloudflare’s insurance-claims demo a Clairic → Clef → GLM-5.3 pipeline ran in 19.3 s at $6,032 per million claims, against 61.7 s and $11,740 for GLM-5.3 at every step.
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- 3 of 5
- Last verified
- 20 October 2026
Your AI and this story
- GPT-6 Astra172 days after its cutoff
- Claude Opus 5.5111 days after its cutoff
- Gemini 4 Argon202 days after its cutoff
- Grok 4.7141 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 111 days before it.
Key facts
- Access: preview on Workers AI as @cf/cloudflare/clairic (env.AI.run() in a Worker or the REST API); billed for input tokens only, rate not stated
- Four tasks: extract values, compute values from the input, estimate values, and forecast future values
- Outputs: typed values, character-offset spans, a probability that the field is null, and a confidence score
- Claims demo (Cloudflare): Clairic → Clef → GLM-5.3 took 19.3 s end to end vs 61.7 s for GLM-5.3 at every step; $6,032 vs $11,740 per million claims; Clairic’s repair estimate was identical across runs while GLM-5.3’s varied; Gemma 4 and Qwen reportedly failed the all-LLM version
- Forecasting demo: TSA checkpoint counts forecast with 3.3% average error; 17 of 21 actual counts fell inside the 80% interval
- No parameter count, architecture details or standard benchmark scores published; open weights planned with no date
- A launch post by @michellechen on X (~183k views) called it ‘a brand new model paradigm’
What happened
Cloudflare added a second non-chat model family to Workers AI, after its Clef decision models of Oct 1. Clairic is built for pulling fields out of documents, emails or forms and returning numbers with calibrated confidence, rather than writing text. Because it produces no output tokens, Cloudflare charges only for input tokens.
The comparisons come from Cloudflare’s own demos, not from standard benchmarks.
Why it matters
It is a bet that many enterprise “LLM” jobs (extraction, scoring, simple forecasting) are better done by fast, deterministic models that report uncertainty, with a general LLM called only for the parts that need reasoning.
Sources
3 sources from 2 sites. Numbers match the chips in the text.
3 sources: 2 primary, 1 reaction
Primary
- Cloudflare blog: Clairic, a System One extraction modelblog.cloudflare.com, official
- Cloudflare on X: Introducing Clairicx.com, official
Reactions
- Michelle Chen on X: ‘a brand new model paradigm’x.com, discussion
Changes
- Filed (Techmeme; Cloudflare blog and X posts read)