Model overview
What this model is for
Clef is a 27B multimodal decision model from Cloudflare, post-trained from Qwen3.8-27B. It accepts text, JSON, images, and video as state, then scores schema-defined choices, yes/no answers, and ordered-score responses as probabilities. This makes it relevant for typed decisions, coding workflows, and vision-driven applications where calibrated choice outputs are more useful than free-form chat. The model exposes a Jev/SystemOne-compatible interface and is not presented as a conventional chat-generation model. Serving requires model-specific code rather than a standard chat endpoint, which may increase integration effort. The open weights use the Apache-2.0 license. The model has 27,356,728,560 parameters. No inference speed or tokens-per-second benchmark is disclosed. The supplied Hugging Face page identifies the Cloudflare/clef repository, but the captured page content does not expose a dated release or detailed freshness information, so source freshness is not disclosed. Developers should verify the current model card and implementation instructions before deployment.