Cloudflare Introduces Clef-omni for Audio, Video, Image and Text Decisions
Clef-omni expands Cloudflare's decision-model family to native audio and video processing alongside images and text.
Cloudflare introduced Clef-omni on October 9, 2026, extending its Clef family of decision models to audio and video alongside text and images.
Decision models emphasize classification and structured judgments rather than producing long-form text. Potential tasks include content categorization, routing requests and checking whether inputs meet defined conditions.
Native multimodal processing may reduce the need to transcribe audio or split video into frames before making a decision, though implementation details and performance vary by workload.
Cloudflare highlighted an example in which a 21-second video with audio was evaluated in roughly 1.5 seconds. This is an illustrative result, not a universal latency guarantee.
The company also reported up to a twofold improvement in Clef inference speed and lower pricing for Clef-flash, changes that can matter for high-volume decision pipelines.
Practical uses include video classification and workflow routing. Safety-critical decisions require additional safeguards and should not rely on a model output alone.
Evaluation must account for false positives, false negatives, language differences and variations in video quality. Aggregate accuracy may hide important failure modes.
The broader significance is the emergence of specialized AI components that decide what action a system should take, rather than merely generating a response.