Overview:
-
Clef-omni is a 30B mixture-of-experts model featuring audio input capabilities, whereas Clef is a dense 27B model.
-
Instead of generating free-form text, both models return probabilities for typed questions.
-
Clef-omni is priced at USD 0.15 per million input tokens, while Clef-flash serves as the budget-friendly option.
Cloudflare’s Clef models focus on making decisions rather than writing essays or chatting. Following the October 9, 2026 rollout of Clef-omni on Workers AI, it joined Clef and Clef-flash within the same family of decision-making models. This raises the question of how Clef-omni and Clef differ.
Essentially, Clef-omni is a mixture-of-experts model capable of reading audio, while Clef is a dense 27B model tailored for precision. Organizations evaluating voice or sound should opt for Clef-omni, whereas those working exclusively with text, images, and video can rely on Clef or the more affordable Clef-flash.
Also read: Google Gives 200 European Researchers Access to Advanced AI Models
Clef-omni & Clef Use Different Architectures
The Clef-omni model utilizes a 30B mixture-of-experts backbone that keeps 3B parameters active concurrently. Cloudflare post-trained this model from Qwen3-Omni-30B-A3B-Instruct, incorporating a joint schema head designed to score every option across all questions simultaneously. This architecture allows the model to maintain extensive knowledge while keeping per-request computational effort comparable to a smaller model.
By contrast, Clef employs a different approach as a 27-billion-parameter model post-trained from a Qwen base, with Clef-flash operating as its smaller 9B counterpart. Clef-omni is distributed under the Apache-2.0 license.
Audio Is the Biggest Input Difference
While Clef-omni accepts state inputs as text, JSON, images, audio, or video, Clef handles text, JSON, images, and video without supporting audio paths. This distinction is critical for applications involving voice chats, call recordings, and sound clips.
Cloudflare’s official changelog specifies WAV and MP3 formats for audio, alongside MP4 and WebM for video. Without Clef-omni, teams moderating voice notes would require an auxiliary speech model paired with custom integration code; with it, a single model processes the entire upload.
Both Models Return Probabilities, Not Prose
Both models operate through an identical interface. Developers transmit a state alongside a schema containing typed questions, and the system responds with a probability for every permitted option in one forward pass, avoiding free-form text generation and output parsing.
As an illustration on Workers AI, a support request can be evaluated to determine urgency and appropriate routing. Because Clef supports between 1 and 64 questions per request and does not bill for output tokens, predicting costs during high-volume operations is simplified.
Also read: Meta’s Muse Wonder AI Challenges ChatGPT and Claude: What Makes It Different?
Pricing and Context Favor Clef-flash for Simple Jobs
On Workers AI, Clef-omni is available for USD 0.15 per million input tokens and features a 64,000-token context window. According to a developer tool listing, Clef is priced at USD 0.24 and Clef-flash at USD 0.09 per million input tokens, with zero charges for output. Because Cloudflare reduced Clef-flash’s pricing and accelerated Clef on October 9, verifying current rates prior to budgeting is recommended.
Clef-omni occupies a mid-tier price point while remaining the sole model among the three to process audio. The same tool recommends Clef for high precision and Clef-flash when prioritizing latency and cost efficiency.
Also read: Anthropic Warns AI Models Could Soon Replicate Across Smart Home Devices
Match the Model to the Moderation Job
Cloudflare highlights primary use cases such as classification, routing, scoring, and guardrails. For instance, a forum dealing solely in text and images can operate using Clef-flash, a marketplace reviewing image and video listings can leverage Clef for enhanced precision, and a gaming or social application featuring voice chat requires Clef-omni.
Because all three models rely on the System One API, teams can begin with smaller options and scale up later without requiring a system rewrite.
Final Thoughts
Although Clef-omni and Clef approach the same task differently, Clef-omni introduces audio capabilities, a mixture-of-experts framework, and intermediate pricing. Meanwhile, Clef provides a dense 27B model optimized for precision across text, images, and video. Users should select Clef-omni when working with audio or mixed media, or choose Clef or Clef-flash for streamlined tasks where speed and cost are paramount.
FAQs
1. What is Cloudflare Clef-omni?
It is a 30-billion-parameter multimodal decision model hosted on Cloudflare Workers AI and released on October 9, 2026.
tech-insider
2. How does Clef-omni differ from Clef?
Clef-omni is built on a 30B mixture-of-experts framework with audio input support, whereas Clef is a dense 27B model lacking audio input.
3. Do these models generate text?
No, they output probabilities for each permitted option rather than producing free-form text.
huggingface
4. How much does Clef-omni cost?
It is priced at USD 0.15 per million input tokens on Workers AI.
cloudflare
5. Which model should I use for voice moderation?
Clef-omni is the appropriate choice, as it is the only model in the pair capable of reading audio.
