
Cloudflare Clef: What Is the New AI Decision Model?Artificial intelligence is moving beyond traditional chatbots and text-generation tools. As AI agents become
Get a free consultation from our team. We build websites, mobile apps, AI chatbots, and custom software for businesses across India.
Stay in the loop with everything you need to know.
Artificial intelligence is moving beyond traditional chatbots and text-generation tools. As AI agents become more common in business applications, developers increasingly need models that can make fast, structured decisions instead of generating long responses.
Cloudflare has introduced a new approach to this problem with Clef and Clef-flash, two decision models launched on October 1, 2026. Available through Cloudflare Workers AI, these models are designed to make fast classifications and decisions while returning probabilities that applications can directly use.
Cloudflare Clef is a decision model, rather than a traditional conversational AI model.
Instead of asking an AI to generate a paragraph and then trying to understand that response in software, developers can give Clef an input and a set of predefined questions or choices.
For example, a customer support system could send a support request to Clef and ask:
Clef can then return structured answers with probabilities. The application can use those results to automatically route a ticket, escalate an issue, or send the case to a human.
Cloudflare explains this new approach in its official announcement about Clef decision models.
Cloudflare launched two models: Clef and Clef-flash.
Clef is the larger model designed for higher-quality decision-making, while Clef-flash is optimized for speed and latency-sensitive applications.
Both models are available through Workers AI, while their open-weight versions are also available through Cloudflare's Hugging Face collection. This gives developers an opportunity to experiment with decision-model workflows and integrate them into their own AI applications.
Traditional large language models are generally designed to understand prompts and generate open-ended text.
Clef works differently.
Its purpose is to make bounded decisions. Instead of generating a long explanation such as "This appears to be a technical support issue because...", it can return a structured choice such as:
Team: Technical — 94% probability
This approach can make AI-agent workflows easier for software to process because the output is already structured.
Cloudflare says its decision models can help agents determine whether they should take an action, which tool they should use, or whether a task should be escalated to a human.
One of Clef's main features is its ability to return probabilities with decisions.
For example, a business application could classify a website as an e-commerce site with a certain probability or identify whether a support request is urgent.
This gives developers a measurable signal that can be used to create rules such as "automate the action only when confidence is above 90%."
Clef also includes a vision encoder, meaning it can process images for classification tasks.
This can make the model useful for applications where decisions depend on both text and visual information, such as website classification, image-based workflows, and AI-powered business automation.
Clef supports a 64K context window, allowing developers to provide more information to the model when making a decision.
This can be useful for AI agents that need to consider large amounts of context before deciding what action to take.
Speed is another major focus of Clef-flash.
Cloudflare's published results show median latency of around 209 ms for Clef and 39 ms for Clef-flash in its evaluations.
For AI agents, this matters because a decision model may sit directly inside a workflow before another action is executed.
Customer Support: Automatically identify urgent tickets and route them to the correct department.
E-commerce: Classify customer requests, detect purchase intent, or decide which workflow should handle an order-related issue.
Security: Classify websites, identify suspicious activity, and help determine whether an event requires further investigation.
AI Agents: Decide which tool an agent should use or whether it should continue, stop, or ask for human assistance.
Business Automation: Convert unstructured business information into structured decisions that software can act upon.
Cloudflare has also discussed using Clef with its Threat Intelligence team to classify website domains. This shows how a specialized decision model can become part of an existing business workflow instead of replacing the entire AI system.
The launch of Clef highlights a broader change in AI development.
Not every AI task requires a model to write an essay, generate code, or have a conversation. Many business workflows simply require an AI system to answer a structured question quickly and reliably.
For example, an AI agent may need to decide:
Should I call this API?
Which department should receive this request?
Is this transaction suspicious?
Should this task be escalated to a human?
For these types of tasks, a specialized decision model could potentially be faster and easier to integrate than using a large generative AI model for every decision.
Cloudflare Clef represents a different direction for AI models: AI that focuses on making decisions rather than generating text.
With Clef and Clef-flash, Cloudflare is targeting developers building AI agents, automation systems, security tools, and business applications where speed, structured outputs, and probability-based decisions are important.
As AI agents become more capable, specialized models like Clef could become an important layer between an AI system's reasoning and the actions that software ultimately performs.
Need help with your digital project? Get in touch with Tech Assistant — we build solutions that scale.