Cloudflare Clef-omni adds audio and video; Clef-flash now $0.038 per million input tokens

Cloudflare released Clef-omni on October 9, an open-weight decision model that takes audio (WAV or MP3) and video (MP4 or WebM) alongside text and images. It is built on Qwen3-Omni-30B-A3B-Instruct, a mixture-of-experts model with 30 billion total parameters and 3 billion active, according to Cloudflare's changelog. Hosted pricing is $0.150 per million input tokens with a 64,000-token context window.

Two other changes shipped with it. Clef-flash fell from $0.090 to $0.038 per million input tokens, but its hosted context window shrank from 64,000 to 24,000 tokens. Cloudflare says only 0.24% of requests exceed that, and the unchanged Hugging Face weights still support 256,000 tokens if you self-host. Hosted Clef is up to 2.0x faster at the median, which Cloudflare attributes to serving-layer work, not new weights.

For context, Cloudflare says the original Clef was conceived on a Friday evening, trained over the weekend and launched the following Thursday.

Now the footnote. Cloudflare's blog puts Clef-omni at about 130 ms for text and about 1.5 seconds for a 21-second video clip with sound. Its changelog says about 20 ms and about 300 ms. Those reported video times differ fivefold, and nothing in either post explains the gap. The benchmark table (98.2 on BFCL case exact, 73.2 on PhishNChips accuracy) is Cloudflare's own, and the announcement includes no independent evaluation. Pick a latency, then show me the eval.