Cloudflare’s Clef-omni adds audio and video input, Clef gets faster and Clef-flash cheaper
Cloudflare launches Clef-omni, an open-weight decision model that takes audio, video, image and text input, and cuts Clef-flash pricing below Jev.
Cloudflare released Clef-omni, a new open-weight decision model that accepts audio, video, image and text input, the company announced on October 9, 2026. The launch follows last week’s debut of Clef and Clef-flash, Cloudflare’s open-weight decision models, and comes alongside a price cut for Clef-flash and speed improvements to Clef.
Clef-omni is built on a Qwen3-Omni-30B-A3B-Instruct mixture-of-experts foundation, according to Cloudflare. The company says it adopts the model’s primary comprehension backbone while discarding the text-to-speech output components. In production, Clef-omni runs a quick prefill pass across the complete payload, scoring all modalities and valid parameter options at once.
One model instead of a pipeline
Until now, Cloudflare says decision models have been largely text-only since the debut of Jev from TypeSafe. Clef supported images and video frame arrays, but Clef-omni adds audio in wav or mp3 and video in mp4 or webm formats alongside text and images.
The company frames this as a way to avoid setting up cascading pipelines of models that transcribe speech-to-text or split audio and image channels from video. Instead, developers can call one model to make decisions across any modality.
Because Clef models are not large language models, they skip output token generation entirely. Cloudflare says this removes the overhead of transcribing or captioning incoming files.
How to call it
Clef-omni is available through Cloudflare’s API under the model identifier @cf/cloudflare/clef-omni. A single request can carry images, audio and video as base64 data URIs, with questions defined as structured parameters. Cloudflare’s example shows a request that asks whether a label is visible in a photo, whether a unit sounds like it is running smoothly, and whether a fan is running in a video.
The model’s open weights are published on HuggingFace, and developer documentation is available on Cloudflare’s site.
Faster Clef, cheaper Clef-flash
Cloudflare also said it made Clef faster and cut the price of Clef-flash so it is now cheaper than Jev. The company did not disclose specific performance figures or pricing in the announcement.
The Clef family came together in under a week, according to Cloudflare: the company decided on a Friday evening to work in the decision model space, trained the model over the weekend, and launched it on Thursday. Cloudflare describes the Clef work as the beginning of continued iteration, with the company saying it will keep pushing to get better, faster and cheaper.
If you want to try Clef-omni, the model is live on Cloudflare’s API and its weights are downloadable from HuggingFace. Check Cloudflare’s developer docs for the full request format and supported input types.
Related reading
Sources
- Introducing Clef-omni with full multimodality, plus a faster Clef and a cheaper Clef-flash — Cloudflare Blog
Some links on this page may be affiliate links. If you buy through them we may earn a commission at no extra cost to you. See our affiliate disclosure.
