tlder@dev — Cloudflare's Clef-omni adds audio and video to its decision model
tlder@dev:~$
AI/ML/Models

Cloudflare's Clef-omni adds audio and video to its decision model

  • Shipped

Clef-omni is the multimodal cut of the decision model Cloudflare shipped on Workers AI at the start of the month. It takes audio and video alongside text and images — WAV or MP3, MP4 or WebM — and runs on a Qwen3-Omni mixture-of-experts base. The weights are on Hugging Face under Apache 2.0, and the hosted version lists at $0.15 per million input tokens. Pricing is the quieter story. Clef-flash dropped to $0.038 per million input tokens, roughly a fifth of the standard model, which holds at $0.24. If you route a lot of low-stakes routing or classification calls through a model, the cheaper tier is the one worth a look.