Cloudflare Releases Clef and Clef-flash: Open-Weight Decision Models That Return Typed Probabilities Instead of Text
Cloudflare ships decision models that return structured probabilities, not text — 27B and 9B weights, sub-40ms latency on Workers AI.
Why it matters
A new model class optimized for classification and structured reasoning rather than generation. Open weights and API compatibility could shift how practitioners build classification pipelines and agentic decision-making systems.
The key facts
10 to knowClef (27B) and Clef-flash (9B) open-weight decision models released
Return typed probabilities instead of text output
Jev-API compatible
Accept images as input
Median latency: Clef 209.3 ms, Clef-flash 38.8 ms on Workers AI
Cloudflare Workers AI deployment target
Output: typed probabilities instead of text
Latency: 209.3 ms median (Clef), 38.8 ms median (Clef-flash) on Workers AI
Deployed on Cloudflare Workers AI platform
Source: MarkTechPost (no direct Cloudflare blog link provided; vendor claims not independently verified)
The story so far
Earlier coverage of this storyline
Go to the source
MarkTechPostmarktechpost.com
Publisher excerpt: Cloudflare has released Clef (27B) and Clef-flash (9B), open-weight decision models that return typed probabilities instead of text. They are Jev-API compatible, accept images, and run on Workers AI at 209.3 ms and 38.8 ms median latency.