Cloudflare launches open-weights model Clef-omni for text, images, audio and video

Cloudflare has launched Clef-omni, an open-weights decision model that can handle text, images, audio and video in a single API call. The weights are available on Hugging Face. Unlike earlier Clef models that only took text, images and video frames, the new model also accepts audio and full video files like wav, mp3, mp4 and webm.

The model is built on Qwen3-Omni-30B-A3B-Instruct and focuses on structured decision tasks rather than writing regular text. In tests, it responded in about 130 milliseconds for text and 150 milliseconds for images, and scored a 21-second video with sound in about 1.5 seconds. Developers no longer need to build separate transcription or splitting pipelines.

Pricing also changed: Clef-flash input dropped from $0.09 to $0.038 per million tokens, a cut of about 58%, while its context window shrank from 64k to 24k. Clef-omni costs $0.15 per million input tokens and $0.60 for output.