Overview:
Clef-omni is a 30B mixture-of-experts model that adds audio input. Clef is a dense 27B model.
Both return probabilities for typed questions, not free-form text.
Clef-omni costs USD 0.15 per million input tokens, and Clef-flash is the budget pick.
Cloudflare’s Clef models do not write essays or chat. They make decisions. Cloudflare rolled out Clef-omni on Workers AI on October 9, 2026. It joins Clef and Clef-flash in the same decision-model family. So how do Cloudflare Clef-omni and Clef differ?
In short, Clef-omni is a mixture-of-experts model that reads audio, while Clef is a dense 27B model aimed at precision. Teams that must judge voice or sound should pick Clef-omni. Teams that handle only text, images, and video can stay with Clef or the cheaper Clef-flash.
Also read: Google Gives 200 European Researchers Access to Advanced AI Models
Clef-omni & Clef Use Different Architectures
Clef-omni is built on a 30B mixture-of-experts backbone with 3B parameters active at a time. Cloudflare post-trained it from Qwen3-Omni-30B-A3B-Instruct and added a small joint schema head that scores all options of all questions together. This design keeps the work per request closer to a small model while holding broad knowledge.
Clef takes a different route. It is a 27-billion-parameter model post-trained from a Qwen base. Clef-flash is the small sibling at 9B. Clef-omni is released under the Apache-2.0 license.
Audio Is the Biggest Input Difference
Clef-omni reads the state as text, JSON, images, audio, or video. Clef processes text, JSON, images, and video, with no audio path. That gap matters for voice chat, call recordings, and sound clips.
Cloudflare’s changelog lists WAV and MP3 for audio and MP4 and WebM for video. Without Clef-omni, a team screening voice notes would need a separate speech model and extra code. With it, one model reads the whole upload.
Both Models Return Probabilities, Not Prose
The two models share the same interface. A developer sends a state and a schema of typed questions, and the model returns a probability for every allowed option in a single forward pass. There is no free-form text and no output parsing.
One Workers AI example asks whether a support request is urgent and which team should handle it. Clef accepts 1 to 64 questions per request, and output tokens are not billed. This makes cost easier to predict when volume is high.
Also read: Meta’s Muse Wonder AI Challenges ChatGPT and Claude: What Makes It Different?
Pricing and Context Favor Clef-flash for Simple Jobs
Clef-omni costs USD 0.15 per million input tokens on Workers AI and has a 64,000-token context window. One developer tool lists Clef at USD 0.24 and Clef-flash at USD 0.09 per million input tokens, with nothing charged for output. Cloudflare cut Clef-flash’s price and sped up Clef on October 9, so check live rates before budgeting.
Clef-omni sits in the middle on price, yet it is the only one of the three that handles audio. The same tool pitches Clef for precision and Clef-flash for latency and price.
Also read: Anthropic Warns AI Models Could Soon Replicate Across Smart Home Devices
Match the Model to the Moderation Job
Cloudflare names classification, routing, scoring, and guardrails as the main uses. A forum with text and images can run on Clef-flash. A marketplace that screens image and video listings can use Clef for more precision. A gaming or social app with voice chat needs Clef-omni.
All three models use the same System One API. Teams can start small and move up later without a rebuild.
Final Thoughts
Clef-omni and Clef solve the same problem in different ways. Clef-omni brings audio, a mixture-of-experts design, and mid-range pricing. Clef offers a dense 27B model built for precision on text, images, and video. Pick Clef-omni for audio or mixed media. Pick Clef or Clef-flash for simpler jobs where cost and speed matter most.
FAQs
1. What is Cloudflare Clef-omni?
It is a 30-billion-parameter multimodal decision model on Cloudflare Workers AI, launched on October 9, 2026.
tech-insider
2. How does Clef-omni differ from Clef?
Clef-omni uses a 30B mixture-of-experts design and reads audio. Clef is a dense 27B model with no audio input.
3. Do these models generate text?
No. They return a probability for each allowed option, with no free-form text.
huggingface
4. How much does Clef-omni cost?
It costs USD 0.15 per million input tokens on Workers AI.
cloudflare
5. Which model should I use for voice moderation?
Clef-omni, since it is the only one of the pair that reads audio.