What happened
Cloudflare introduced Clef, a set of open-weight AI models for selecting between predefined options. Instead of producing free-form text, it returns structured choices with associated probabilities.
The line-up includes a larger multimodal model and a smaller version built for decisions where latency matters. The company also says it has made available a platform to fine-tune these models with its own data and adapt them to tasks such as support triage or bot classification.
According to Cloudflare, these models can help AI agents decide the next action, such as routing a request, escalating it, or passing it to a person. They are also available in Workers AI and as downloadable weights on Hugging Face, with developers invited to test them and send feedback.
Key points
- Clef chooses between predefined options instead of generating text.
- The larger version is multimodal and accepts text, JSON, images and video.
- Cloudflare also launched Clef-Flash for lower-latency decisions.
- The company wants to fine-tune the models for tasks such as support triage and bot classification.
- The models are available through Workers AI and on Hugging Face.
Why it matters for your organisation
For companies and public bodies, this type of model can speed up repetitive decisions and reduce the need to interpret free-form text. The value lies in using AI to classify, route and prioritise requests more consistently, while still keeping human intervention available when needed.