Introducing Clef Decision Models
Cloudflare is releasing its open-weight decision models Clef and Clef-flash, trained in-house and hosted on Workers AI. Clef leads the Jev Decision Index benchmark, making it the first model to surpass the Jev System One.
A decision model produces classifications that help agents decide how to act, based on probabilities. It generates bounded, structured outputs for any input set without retraining for new categories — unlike the non-deterministic, open-ended outputs of LLMs.
Key Features and Benefits
Both models are fully open-sourced under Apache 2.0 on Hugging Face for local experimentation. Clef-flash is optimized for low-latency use cases. They are fully Jev-API compatible for easy integration into workflows.
Cloudflare is also launching a new RL fine-tuning platform that allows customers to adapt Clef to their specific use cases. This lowers barriers to deployment and reduces reliance on human-in-the-loop for agentic decisions.
The open-source release and RL fine-tuning platform mark the mainstreaming of decision models, empowering enterprises to build smarter, more reliable AI agents.