Cloudflare

Cloudflare is a global cloud platform that provides content delivery, edge computing, security, and networking services. A significant portion of the internet's traffic flows through Cloudflare's network, making it one of the most critical infrastructure providers in the technology stack.

Cloudflare's strategic positioning in the agentic economy centers on Workers AI—their edge compute platform that enables AI inference to run close to users rather than in centralized data centers. This is a bet that as agents become the primary interface to the internet, the latency of inference will matter enormously. The difference between inference that takes 5 seconds and inference that takes 500 milliseconds is the difference between an agent that feels like a colleague and one that feels like a bureaucracy.

In the Seven Layers of the Agentic Economy, Cloudflare sits at Layer 3: Platforms & Services, providing the edge infrastructure that agentic applications depend on for performance and reliability.

2026 Developments: Clef Decision Models

On October 1, 2026, Cloudflare released Clef and Clef-flash, two open-source decision models, under the Apache 2.0 license on Hugging Face. In the launch post, Cloudflare's Michelle Chen describes the category this way: "A decision model makes classifications to help agents decide how to act, based on certain probabilities." Instead of generating text, Clef scores a fixed set of allowed answers and returns one with a calibrated probability.

How Clef works

Clef is built on a Qwen3.8-27B backbone and Clef-flash on Qwen3.5-9B. Cloudflare says the decision step is "non-autoregressive, so there's no intermediate text to generate token by token"; the model "scores the valid schema choices in parallel" using two-stage attention routing. Training adds LoRA adapters and a routing head, with label-smoothed cross-entropy "paired with a Brier loss to refine probability calibration," plus Reinforcement Learning for Calibrated Decisions (RLCD). The models include a vision encoder and a 64k-token context window.

Cloudflare made Clef "fully Jev-API compatible," matching the interface of TypeSafe's Jev, which launched days earlier. In Cloudflare's own testing, median latency was 209.3 ms for Clef and 38.8 ms for Clef-flash, against 524.1 ms for Jev, and Cloudflare says Clef currently leads the community Jev Decision Index. These are vendor benchmarks and have not been independently reproduced.

Availability and fine-tuning

Both models are generally available on Workers AI. Customers can route their traffic through Cloudflare's AI Gateway to "automatically create a dataset" from real requests, which can then be used to fine-tune a Clef model. For now, fine-tuning goes through Cloudflare's forward-deployed engineers, with a self-serve option planned.

Internal uses

Cloudflare says it already runs Clef in its own products: classifying domains, triaging support requests, evaluating trust and safety submissions, and in its bot products "to decide if a crawler is a good bot or bad bot." Bot classification is a natural fit for the company: it is a high-volume, latency-sensitive yes/no decision made at the edge, which is exactly the workload Workers AI was built for.