In an Oct. 5 post, @ClementDelangue described Cloudflare’s Clef as trending on Hugging Face, four days after its release. His post gives no numeric rank in its text.
Decisions with defined answers
Cloudflare’s Oct. 1 announcement introduced Clef and the smaller Clef-flash on Workers AI. They read a situation and typed questions, then return probabilities for the allowed answers. That lets an application choose a support-ticket destination or decide whether to escalate a request without parsing a paragraph of generated prose.
The models support yes-or-no questions, choices among named options and scores against an ordered rubric. Clef has 27 billion parameters; Clef-flash has 9 billion.
The Clef model card identifies Qwen3.8-27B as its backbone and describes inputs including text, JSON, images and video. It documents compatibility with the Jev/SystemOne API.
Availability and evaluation
Cloudflare offers the models through Workers AI and makes their weights downloadable under Apache 2.0. The model card labels its Decision Index results as an internal company run.