Updated

Cloudflare Clef-omni

Cloudflare released Clef-omni, an open-weight decision model that accepts audio, video, image, and text inputs in one API call. Its release builds on the existing Clef family and makes its weights available on Hugging Face.

By No Code MBA · Facts checked

Provider
Cloudflare
Announcement date
Status at last check
Generally available
Model family
Clef
Preceding model
Clef
Capabilities
Text, Image, Audio, Video

What changed

Clef-omni adds audio and video input to Cloudflare's Clef decision models, taking text, images, audio and video in one call with no transcription or captioning step. It is post-trained with LoRA adapters on Qwen3-Omni-30B-A3B-Instruct and returns probabilities for typed questions instead of generating text. Cloudflare reports median latency of about 130 ms for text, 150 ms for images and about 1.5 seconds for a 21-second video. Cloudflare also cut hosted Clef latency 1.7x to 2x and lowered Clef-flash pricing.

Access and availability

Hosted on Workers AI as @cf/cloudflare/clef-omni with a 64,000 token context window, and as open weights on Hugging Face at Cloudflare/clef-omni under Apache 2.0.

Official access documentation

Pricing

$0.15 per million input tokens (media converted to input tokens); output not charged

Official pricing

What you can build

Try Clef-omni on a workflow that currently chains transcription with image or video analysis, then compare decision quality and latency against that pipeline.

Limitations

Requests accept up to 4 images, 4 audio clips (300 seconds each) and 2 videos (60 seconds each), media must be embedded since remote URLs are not accepted, and the model cannot generate free-form text. Cloudflare's own benchmarks show Clef-omni trailing Clef on several text tasks such as When2Call and home appliances. Self-hosting needs about 64 GB of GPU memory.

Sources and release history