Cloudflare launches open-weight Clef decision models on Workers AI and an RL fine-tuning platform for structured AI workflows

Cloudflare released Clef and Clef-flash, two hosted decision models designed to produce fast, low-cost, structured outputs for workflow routing and agent decisions. The models are Jev API-compatible, open-sourced on Hugging Face under Apache 2.0, and can be customized through Cloudflare’s new reinforcement learning fine-tuning product.

The launch reflects growing demand for smaller, deterministic AI components that complement LLMs by handling repeatable classification and routing tasks. Open weights plus hosted deployment could make agentic workflows cheaper and more reliable, especially for businesses that need auditable decisions rather than open-ended generation.