Cloudflare has introduced two new AI models, Clef and Clef-flash, which are designed to quickly make specific decisions: answer "yes" or "no", choose one of the options, or prioritize them. The company positions them as a competitor to the Jev model from TypeSafe AI.
Cloudflare has introduced two new AI models, Clef and Clef-flash, which are designed to quickly make specific decisions: answer "yes" or "no", choose one of the options, or prioritize them. The company positions them as a competitor to the Jev model from TypeSafe AI.
This is reported by Cloudflare.
Clef can be useful where AI needs to quickly decide what to do next. For example, determining whether a customer’s request is urgent, who to transfer it to, or what task an agent should perform next. With a single query, the model can answer dozens of such questions at once, simultaneously evaluating all available options.
Cloudflare also added a feature that Jev lacks. Clef can work with images and videos. That is, decisions can be made based on, for example, a camera frame or other visual content.
There are two versions of the model. The full Clef is based on Qwen3.8-27B, and the lighter and faster Clef-flash is based on Qwen3.5-9B. Both support a context window of up to 64,000 tokens. Cloudflare also claims that Clef outperforms Jev in speed and accuracy in a number of tests. According to the company's own testing, Clef's median response time was about 209 ms, compared to Jev's 524 ms, while Clef-flash only had about 39 ms.
Clef can be run through Workers AI, Cloudflare’s cloud platform that allows developers to run AI models on the company’s infrastructure without their own servers or GPUs. Clef costs $0.24 per 1 million input tokens to use, while Clef-flash costs $0.09. For comparison, TypeSafe has set a price for Jev at $0.042 per 1 million input tokens.
Cloudflare has also published the parameters of the Clef models on Hugging Face under the Apache 2.0 license, so they can be run locally. According to The Register, Clef-flash will require a GPU with at least 41 GB of video memory, and full Clef will require about 85 GB, assuming a single concurrent request and using a 64,000-token context window.
Clef is also compatible with the Jev API, so developers who already use the TypeSafe model can replace Jev with Clef in their services or applications without significantly reworking their already configured integration.
Previously, dev.ua talked about how Ukrainian IT professionals tested Jev in real products and how they evaluated the model.
