Cloudflare puts audio and video into its decision model with Clef-Omni (runtimewire.com)

Cloudflare has announced the launch of Clef-omni, an advanced decision model now capable of processing audio and video inputs alongside text and images in a single API call. This extension aims to streamline the decision-making process by eliminating the need for separate media-processing steps, thus enhancing efficiency. Clef-omni is built on a 30B-A3B mixture-of-experts architecture derived from the Qwen3-Omni-30B-A3B-Instruct model, integrating specialized encoders for audio and vision, along with a joint schema head that evaluates responses across various media types. This release is significant for the AI/ML community as it positions Cloudflare directly against competitors like TypeSafe in the burgeoning area of multimodal decision-making. Although Clef-omni offers substantial cost savings and faster response times—reportedly around 130 to 150 milliseconds for text and images, and approximately 1.5 seconds for video—it also comes with tradeoffs, such as a reduced context window for hosted models. Cloudflare’s approach illustrates a strategic emphasis on making multimodal processing practical for various applications, though the long-term viability and accuracy of the model will hinge on independent assessments of its effectiveness in real-world scenarios.