Clef Omni

Clef Omni is the member of Cloudflare's open Clef family of decision-only AIs that reads not just text but also images, audio and video, returning a yes/no answer, a choice, or a score with a probability instead of writing text. It was retrained from an Alibaba Qwen model that handles many formats; the whole model is large but uses only a small part at a time, which keeps it fast, and anyone can download and run it themselves under the Apache license. In our real-time game tests its quick answers put it near the top in fast-reaction games like Whack-a-Mole and Snake, but it was weak in Tetris and slightly less accurate than Clef on our Korean and Japanese judgment tests. Cloudflare itself notes that it scores below the text-only Clef on some tests; running it yourself needs a large amount of GPU memory, and since it cannot write text it cannot transcribe or caption anything.