Clef is available to run locally and in the Ollama cloud
Ollama has made the Clef model from Cloudflare available to run locally and in the cloud. The model has 27 billion parameters, accepts text and images, and returns structured decisions with probabilities. It is released under the Apache 2.0 license.
The Clef model from Cloudflare is now available to run locally and in the Ollama cloud. It has 27 billion parameters, was fine-tuned from Qwen3.8-27B and is released under the Apache 2.0 license. It accepts both text and image inputs and is compatible with the Jev and System One APIs.
The user provides an input scenario and a schema of questions with specified data types. The model returns structured decisions and probabilities for individual answers. According to the Ollama blog, it evaluates all answers in a single pass through the input. Listed uses include request routing, rule evaluation, working with photographs of documents and checking tool calls made by an agent before execution.
The blog reproduces measurement results from Cloudflare. Across 43 evaluation tasks, it reports a median latency of 209.3 ms and a 95th percentile of 238.6 ms for Clef. The published comparisons show differences depending on the task: for example, on BANKING77, Clef achieved a macro-F1 score of 94.20 compared with 79.74 for Jev, while on When2Call it had lower accuracy.
Why it matters
A structured decision with probabilities can be used to automatically select the next step based on text or an image. Availability both locally and in the Ollama cloud makes it possible to try the same model in both environments; however, the published measurements alone do not determine how well it will perform in a specific workflow.
Two audiences, two different impacts
What this means
For individuals
When working with photographs of forms, individuals can try evaluating individual fields using predefined questions and receive probabilities alongside the answers.
For a business
For business processes, the model is available for routing customer requests or checking tool calls made by an agent before execution. The source lists these uses but does not establish their reliability within a specific company.
ProcessesCheck the original
Event sources
only one source so far · 1 publisher, 0 independent. We count feeds from the same owner only once.