Cloudflare · 2026-10-09 · major
Clef-omni — Cloudflare's open decision model now reads audio and video
Clef-omni is Cloudflare's new Apache-2.0 decision model on Qwen3-Omni-30B-A3B. It scores typed questions about text, images, audio and video in one call. Cloudflare also made Clef up to 2x faster and cut Clef-flash to $0.038 per 1M tokens.
Cloudflare's Clef-omni answers yes/no and multiple-choice questions about audio, video, images and text in one call.
Key specs
| Text decision latency (median) | ~130 ms |
|---|---|
| Clinc150+oos (macro f1) | 97.7 |
Quick facts
| Maker | Cloudflare |
|---|---|
| Base model | Qwen3-Omni-30B-A3B-Instruct (MoE) |
| Inputs | Text, JSON, images, audio, video |
| License | Apache-2.0 |
| Context window | 64K tokens on Workers AI |
| Model ID | @cf/cloudflare/clef-omni |
Pricing
| Clef-omni input | $0.15 / 1M tokens |
|---|---|
| Clef input | $0.24 / 1M tokens |
| Clef-flash input · was $0.09 | $0.038 / 1M tokens |
| Output tokens · Clef models do not charge for output | $0 |
What is it?
Clef-omni adds audio (WAV, MP3) and video (MP4, WebM) to Cloudflare's Clef decision models, released on October 9, 2026. Like Clef and Clef-flash, it does not write text: it takes an input plus typed questions and returns a calibrated probability for each allowed answer, through the Jev-compatible API.
How does it work?
Under the hood is Qwen3-Omni-30B-A3B-Instruct, a mixture-of-experts model, with its text-to-speech output parts removed. Cloudflare keeps that backbone frozen and trains LoRA adapters with label-smoothed cross-entropy plus a Brier score loss, so the scores behave like real probabilities.
Why does it matter?
Agents that route calls, check uploads or triage screen recordings can now make a typed decision on raw media without a separate transcription or captioning step. The same launch cuts Clef-flash's price by more than half, making cheap high-volume routing on Workers AI cheaper still.
Who is it for?
developers building agent routing, moderation and triage on media
Frequently asked questions
- How fast is Clef-omni on audio and video?
- Cloudflare reports that Clef-omni answers text-only decisions in about 130 ms at the median and image decisions in about 150 ms. Audio clips take a few hundred milliseconds, and a 21-second video clip with sound takes about 1.5 seconds. Workers AI accepts up to two embedded video clips of up to 60 seconds each per request.
- How does Clef-omni compare with Clef and Jev?
- In Cloudflare's launch table Clef-omni has the best Clef-family scores on BANKING77 (94.8), CLINC150+OOS (97.7) and Amazon ESCI (57.8), ahead of Jev's 79.74, 89.27 and 55.21. It trails Clef and Clef-flash on the home-appliances task (69.3 against 82.95 and 97.73) and Jev on When2Call (63.3 against 80.97).
- What changed for existing Clef and Clef-flash users?
- Cloudflare moved hosted Clef to SGLang, which makes the same weights 1.7x to 2.0x faster at the median; ~800-token inputs dropped from 262 ms to 152 ms. Clef-flash fell from $0.09 to $0.038 per million input tokens, but its hosted context window shrank from 64K to 24K tokens. Cloudflare says only 0.24% of requests were longer than that.
- Can I run Clef-omni on my own GPUs?
- Yes. Cloudflare published the Clef-omni weights on Hugging Face as Cloudflare/clef-omni under Apache 2.0, and the model card includes Python code that loads the model and prints a probability for each option. Cloudflare's SGLang changes for Clef are also upstream in SGLang 0.5.22.
Try it
@cf/cloudflare/clef-omni on Workers AI