While the tech industry continues to burn millions on stochastic parrots and open-ended chatbots that hallucinate their way through basic workflows, Cloudflare has quietly decided to cut through the noise. As noted by the Cloudflare engineering team, the company is rolling out two trained decision models—Clef and Clef-flash—hosted directly on Workers AI, offering a pragmatic alternative for structured business decisions without the endless cycle of retraining classifiers.
Unlike traditional Large Language Models that love to write poetry when you just need a boolean, these models are designed for deterministic outputs. According to Cloudflare's technical documentation, Clef models are now open-sourcing on Hugging Face under an Apache 2.0 license, allowing developers to run them locally rather than renting expensive cloud inference by the token.
"Lastly, we’re excited to debut our new reinforcement learning (RL) product, which allows customers to fine-tune Clef to suit their use cases as well."
Cloudflare's new reinforcement learning product essentially hands the steering wheel back to the developers. This setup allows automated workflows to evaluate incoming payloads, return strictly typed answers complete with probability scores, and trigger downstream programmatic actions entirely without human babysitting or unpredictable agentic loops.
Performance Benchmarks and Latency
Internal operational telemetry from Cloudflare makes for an uncomfortable read for generative AI maximalists. When the Cloudflare Threat Intelligence team deployed the Clef model for domain classification tasks, the drop in latency and the absence of creative hallucinations exposed just how much compute we have been wasting on glorified autocomplete.
If your current routing workflows on Workers AI are still relying on open-ended LLM endpoints just to parse structured JSON, it is time to look at the math. Swapping out bloated generative models for deterministic decision metrics isn't just about shaving off milliseconds—it is about keeping your inference budget from turning into a corporate bonfire.