Cloudflare Clef: 39ms AI Decision Models for Autonomous Agents

According to Cloudflare, the newly launched Clef and Clef-flash decision models eliminate the need for human intervention in routine AI agent tasks by returning structured probabilities instead of raw text. Built on Qwen architectures, these models assign likelihoods to predefined options, allowing downstream systems to execute decisions instantly.

How fast are Cloudflare's Clef models in practice?

Cloudflare reports a median response time of 39 milliseconds for Clef-flash and 209 milliseconds for the larger Clef model. In comparative tests, this performance outpaces TypeSafe AI's Jev model, which logs a median latency of over 524 milliseconds.

What changes for engineering teams on Monday morning?

Engineering teams can route support tickets, filter abuse reports, or classify incoming data streams without waiting for slow general-purpose language models. Processing an entire website classification task takes 2.2 seconds with Clef, compared to 4.7 seconds for standard LLMs, cutting wait times in half for automated pipelines.

Both models run directly on Cloudflare Workers AI and are available under the Apache 2.0 license, allowing developers to integrate structured decision-making into existing edge applications immediately.