Cloudflare launches open-weight decision AI model Clef-omni with audio and video support

Cloudflare has launched the open-weight decision AI model Clef-omni, which supports processing text, images, audio, and video through a single API call. The model extends the capabilities of the previous Clef series by adding full video and audio input support, including formats like wav, mp3, mp4, and webm. Developers can now handle multiple input types without additional preprocessing steps. Clef-omni is built on Qwen3-Omni-30B-A3B-Instruct and is designed for structured decision tasks rather than generating standard text. In tests, it achieved a median response time of 130 milliseconds for text, 150 milliseconds for images, and 1.5 seconds for a 21-second video with audio. Cloudflare also reduced the price of Clef-flash to 0.038 USD per million input tokens, a 58% decrease from 0.09 USD, and reduced the context window from 64k to 24k tokens. The model's weights are now available on Hugging Face, making it accessible to a broader audience. Cloudflare emphasized that Clef-omni simplifies the integration process for developers by eliminating the need for separate audio processing workflows. The company highlighted the model's efficiency and versatility, positioning it as a valuable tool for applications requiring multi-modal input analysis.