Cloudflare Cuts Clef-flash API Input Token Price by 58%, Trims Context Window to 24k

On Oct. 9, Cloudflare announced price cuts for its Clef-flash model API. The price per million input tokens was reduced from $0.09 to $0.038, a drop of about 58%, while the output price is $0.12 per million tokens. The managed context window was also trimmed from 64k to 24k.

The pricing change accompanied the launch of Clef-omni, an open-weight decision model built on Qwen3-Omni-30B-A3B-Instruct. It retains the base model's core understanding capabilities and targets structured decision-making tasks rather than general text generation. Clef-omni adds support for audio and full video input, handling formats such as wav, mp3, mp4 and webm, so developers do not need to build separate transcription or audio/video splitting pipelines. Earlier Clef models already accepted text, images, and continuous frames sampled from video.

According to Cloudflare's test data, median response times were about 130 ms for plain text requests, 150 ms for image requests, and about 1.5 seconds for a 21-second video with audio. The model weights are available on Hugging Face.