Clef-omni
AI model · Multimodal model · Text → Text by Cloudflare
Multimodal decision model for agentic workflows and high-speed classification.
Clef-omni is a 30B-parameter multimodal decision model from Cloudflare that turns text, audio, video, and images into structured decisions. It is ideal for developers building agentic workflows that require high-speed, multi-format state analysis at the edge.
Model specs
- Provider
- Cloudflare
- Model family
- Clef
- Version
- omni
- Released
- 2026-10-09
- Input
- Text, Image, Audio, Video
- Output
- Text
- Parameters
- 30B (3B active)
- Capabilities
- Vision, Structured output
- Launched
- 2026-10-09
- Languages
- English
Category: Generative AI
- Parameters
- 30B (3B active)
- Sizes and variants
- Clef, Clef-flash, Clef-omni
- Access
- Workers AI, REST API
- Released
- 2026-10-09
- Backbone Parameters
- 30B
- Active Parameters
- 3B
- 30B-parameter mixture-of-experts backbone
- 3B active parameters for efficient inference
- Natively processes audio, video, images, and text
- Turns complex states into typed decisions with probabilities
Overview
Clef-omni is a multimodal decision model developed by Cloudflare, designed for businesses and developers to automate complex decision-making tasks. Built on a 30B-parameter mixture-of-experts (MoE) backbone with 3B active parameters, it natively processes audio, video, images, and text in a single pipeline. The model solves the problem of turning unstructured or multi-format data into structured, typed decisions by reading input states and returning probabilities for allowed options defined in a question schema. It is delivered through Cloudflare's Workers AI platform, allowing it to run at the edge for high-speed classification and agentic workflows. Clef-omni sets itself apart by providing full multimodality and efficient inference within the Clef family of decision models.
Multimodal Intelligence
- Full Multimodality: The model natively processes audio, video, images, and text in a single pipeline for unified decision-making.
- Flexible State Input: It reads input states as text, JSON, images, audio, or video to return probabilities for allowed options.
- Decision-Centric Output: Clef-omni turns a state and a schema of typed questions into decisions with associated probabilities.
- MoE Architecture: The model is built on a 30B-parameter mixture-of-experts backbone with 3B active parameters for efficient inference.
Deployment and Integration
- Edge Deployment: The model is hosted on Cloudflare's global network, allowing for high-speed classification and agentic workflows.
- Developer Access: Developers can access the model through Workers AI bindings or the REST API.
- Agentic Workflows: Clef-omni is designed for high-speed classification and complex agentic workflows.
- Typed Decisions: The model returns a probability for every allowed option defined in a question schema.
Use cases
- Developers use Clef-omni to process video and audio feeds to make real-time decisions in automated monitoring systems.
- Businesses integrate the model into agentic workflows to turn complex customer interactions into structured data for downstream processing.
- Security teams leverage the model's multimodality to classify multi-format alerts and determine the probability of successful exploits.
Specifications
- Modalities
- Text, JSON, Images, Audio, Video → Probabilities for typed questions
- Parameters
- 30B (3B active)
- Sizes and variants
- Clef, Clef-flash, Clef-omni
- Access
- Workers AI, REST API
- Released
- 2026-10-09
- Backbone Parameters
- 30B
- Active Parameters
- 3B
- Architecture
- Mixture-of-Experts (MoE)
- Input Modalities
- Text, JSON, Images, Audio, Video
- Output Format
- Probabilities for typed questions
- Platform
- Cloudflare Workers AI
- Task Type
- Text Generation / Decision Making
How it works
- Define Question Schema: Developers define a schema of typed questions and the allowed options for the model to evaluate.
- Provide Input State: The model accepts input states in various formats, including text, JSON, images, audio, or video.
- Execute Inference: The model processes the input through its 30B-parameter MoE pipeline to calculate probabilities.
- Receive Decisions: The model returns a probability for every allowed option, enabling automated decision-making in agentic workflows.
FAQ
How many parameters does Clef-omni have?
It uses a 30B-parameter mixture-of-experts backbone with 3B active parameters.
What input types does Clef-omni support?
The model natively processes audio, video, images, and text in a single pipeline.
How can I access Clef-omni?
Clef-omni is hosted on Cloudflare's Workers AI platform and can be accessed via Workers bindings or the REST API.
Added by AntSpace from public sources (developers.cloudflare.com).