Cloudflare Launches Clef and Clef-flash Models for Autonomous AI Decision-Making

Cloudflare has released its Clef and Clef-flash models, which allow AI agents to make decisions autonomously, eliminating the need for human intervention. Clef utilizes the Qwen3.8-27B base model, while Clef-flash is based on Qwen3.5-9B. Both models employ the Reinforcement Learning for Calibrated Decisions (RLCD) training method, resulting in a median response time of 209 milliseconds for Clef and 39 milliseconds for Clef-flash. In comparison, TypeSafe AI's Jev model has a median response time of 524 milliseconds.

Accuracy metrics show Clef achieving 91.93% accuracy, while Clef-flash slightly outperforms at 93.11%. In contrast, Jev has an accuracy of 88.19%. Clef's context window size is 64,000 tokens, significantly enhancing its decision-making capabilities. The models can classify websites with a probability of 95% for fashion sites and 85% for online stores, while maintaining a phishing site detection rate of under 1%.

Cloudflare's advancements follow its acquisition of Replicate in late 2025, which has enabled the development of custom models. This release positions Cloudflare as a strong competitor against TypeSafe AI's Jev model, particularly in the realm of autonomous AI decision-making. As Cloudflare states, "a human does not necessarily need to be in the loop for agentic decisions anymore," indicating a shift towards greater AI autonomy in operational contexts.