When One Model Handles Every Kind of Input, Pipelines Get a Lot Simpler

Most AI automation tools today are secretly several models working in a chain. One reads text, another transcribes audio, a third describes images. That chain takes time, costs money at every step, and breaks when any one part fails. One model that handles all input types changes the math entirely.

On October 9, Cloudflare shipped Clef-omni, a decision model that accepts text, images, audio files (WAV or MP3), and video clips (MP4 or WebM) in a single API call and returns a structured label or route. It is built on a 30-billion-parameter open model and costs $0.15 per million tokens through Workers AI. Cloudflare also released the model weights on Hugging Face, so any team can run it on their own servers, for free.

The speed numbers are solid. Text decisions come back in about 130 milliseconds, images in 150 milliseconds, and a 21-second video clip with audio in roughly 1.5 seconds. On BFCL — a standard test that checks whether models follow tool-call instructions precisely — Clef-omni scored 98.2%, high enough to trust in production.

This matters most for teams that triage or route incoming work: customer support, content moderation, legal intake, fraud review. Those queues already mix text tickets, screenshots, voice messages, and short video clips. Right now, handling all of them means three or four separate models, three or four API calls, and three or four bills. Clef-omni collapses that into one — and for anyone processing thousands of mixed-media items a day, that simplification adds up fast.

If your product touches documents, voice, or video in any way, this is worth testing now. The weights are on Hugging Face, the API is live on Workers AI, and the price is low enough to evaluate in a real workflow this week. Building the single-model path today is much easier than untangling a chain later.