Cloudflare’s Clef ships 9B and 27B decision models for agents

Cloudflare is treating agent control as a bounded decision problem rather than a text-generation task. Clef delivers open-weight 9B and 27B models plus a platform for adapting them to specific decisions. Its 27B model accepts state and a typed-question schema, then returns probabilities across allowed options in one pass, including for text, JSON, images, and video.

• Cloudflare released Clef in 9B and 27B parameter sizes, alongside a platform for adapting the models to specific decision-making tasks.
• The 27B multimodal model takes a state and schema of typed questions as input, then returns a probability for each permitted option for every question in a single forward pass.
• Clef can process text, JSON, images, or video, giving agent workflows one decision interface across different input types.

Bottom line If your agents choose among bounded actions, test Clef against your current decision logic; otherwise, skim this release.