Cloudflare officially announced on October 9 the launch of its new open-weight decision-making model Clef-omni. This model extends the original text and image processing capabilities to support audio and full video formats, enabling the processing of multimodal content with a single API call.

image.png

Multimodal Upgrade and Architecture Analysis

In terms of multimodal support, the traditional Clef model mainly handled text, images, and static consecutive frames extracted at time intervals. The newly launched Clef-omni, however, natively supports audio and video formats such as wav, mp3, mp4, and webm. This improvement eliminates the cumbersome steps that developers previously needed to deploy for speech transcription and audio-video segmentation, allowing a single model to efficiently handle diverse input requirements.

According to official technical details, Clef-omni is built upon Qwen3-Omni-30B-A3B-Instruct and retains its core understanding capabilities. This model primarily targets structured decision-making tasks and does not generate conventional text outputs. In terms of performance, the median response time for pure text requests is approximately 130 milliseconds, and for images, around 150 milliseconds; processing a 21-second video with sound takes about 1.5 seconds to complete the evaluation.

Price Reduction and Series Model Comparison

With the release of the new model, Cloudflare also reduced the usage price of Clef-flash, lowering the cost from 0.09 USD per million input tokens to 0.038 USD, a reduction of about 58%. The context window of its hosted version was adjusted from 64k to 24k. Overall, the tiered pricing of the current series of models meets the cost and performance needs of different developers.