Cloudflare launches free AI models that answer only with numbers

US tech company Cloudflare on October 1, 2026, released two new AI models: Clef and Clef-flash. They are built for a specific task that many AI systems spend a lot of time on behind the scenes: making small decisions, for example whether a message should be routed to department A or B, or whether a customer's reply is yes or no.

Instead of writing a long answer in text, as you know it from ChatGPT, the Clef models return a number, a probability, for each of the answers you have asked for. For example, you can ask the model whether an email is spam and get an answer like "87 percent probability of yes." That makes the answers faster to use in programs, because the computer doesn't have to interpret text, just read a number.

Two sizes for different needs

Clef is the large model with 27 billion parameters (parameters are, roughly speaking, the model's internal settings, and more parameters usually mean a smarter but slower model). Clef-flash is the small and fast version with 9 billion parameters, built to respond in a few tenths of a second.

According to Cloudflare's own figures, Clef-flash responds in under 40 milliseconds on average, which according to the company is significantly faster than a competing service called Jev. The models can also see images and video, not just text.

Free and open

What makes the news interesting for regular users and businesses is that the models are "open-weight." This means the fully trained model itself can be downloaded for free from the Hugging Face platform and run on your own computers, without having to pay Cloudflare for it. The models are built on top of the Chinese Qwen model, but have been further trained by Cloudflare to specifically make this type of decision.

Cloudflare hopes Clef can be used by other companies building AI agents, meaning AI systems that have to make a series of smaller decisions on their own during a task, without a human having to read and interpret a long text for every single step.

What it means for you

If you are a developer or work with IT in your company, you can use Clef to make your AI systems faster and cheaper for simple tasks like spam filtering or sorting messages. You can download the models for free on Hugging Face and test them on your own data. As a regular user, you will probably feel the effect indirectly, for example in customer service chatbots or spam filters responding faster.