Cloudflare has released Clef and Clef-flash, two decision models designed to return structured classifications for AI workflows. Announced on October 1, the models are hosted on Workers AI, with weights published on Hugging Face under an Apache 2.0 license.
Instead of generating an open-ended response, a decision model selects among defined choices and returns probabilities. Cloudflare gives examples such as deciding how to route a support message or classify a website. Clef also accepts visual inputs and supports a 64,000-token context window, according to the company.
The release includes a hands-on fine-tuning service through Cloudflare’s engineering team. A broader self-service reinforcement learning platform is described as a future development, so customers should distinguish the available service from the planned tooling.
For application builders, these models offer a focused component between collecting information and taking an action. A fast classification can reduce workflow overhead, but its usefulness depends on the choices, examples and thresholds supplied by the developer. Cloudflare’s benchmark claims come from its own evaluations and should not be treated as universal results for every task.
