Clef-flash
AI model · Multimodal model · Text → Text by Cloudflare
Fast multimodal decision model for high-speed agentic workflows.
A fast 9B-parameter multimodal decision model for high-speed classification and agentic workflows. It converts text, image, and video state data into structured decisions with probabilities, running natively on Cloudflare's global edge network.
Model specs
- Provider
- Cloudflare
- Model family
- Clef
- Released
- 2026-10-01
- Input
- Text, Image, Video
- Output
- Text
- Weights
- Open weights
- License
- Open-source
- Parameters
- 9B
- Capabilities
- Vision, Structured output
- Available on
- Workers AI
- Launched
- 2026-10-01
- Languages
- English
Category: Machine Learning
- Modalities
- Text, JSON, Images, Video → Decisions with probabilities
- Parameters
- 9B
- Sizes and variants
- Clef (27B), Clef-flash (9B), Clef-omni (30B MoE)
- License
- Open-source decision models hosted on Workers AI
- Access
- Workers AI (Workers Bindings, REST API)
- Released
- 2026-10-01
- 9B-parameter multimodal decision model
- Supports text, JSON, image, and video inputs
- Optimized for high-speed classification and agentic workflows
- Returns probabilities for allowed options in a typed schema
- Hosted on Cloudflare's global edge network via Workers AI
Overview
Clef-flash is a 9B-parameter multimodal decision model developed by Cloudflare and hosted on the Workers AI platform. It is designed for businesses and developers who need to turn unstructured state data—including text, JSON, images, and video—into structured, deterministic decisions. The model works by taking a state input and a schema of typed questions, returning a probability for every allowed option defined in the schema. As part of the Cloudflare developer platform, it runs natively on a global edge network, allowing for high-performance inference close to users and data.
Decision Intelligence
- Multimodal Input Support: Clef-flash processes unstructured state data provided as text, JSON, images, or video.
- Schema-Based Decisions: The model turns a state and a schema of typed questions into structured decisions.
- Probabilistic Outputs: It returns a probability for every allowed option of every question in the schema.
- High-Speed Inference: Clef-flash is optimized for high-speed classification and low-latency agentic workflows.
Developer Platform Features
- Edge-Native Deployment: The model is hosted on Cloudflare's global network for edge-native execution.
- Flexible Access Methods: Developers can access the model through Workers AI using Workers Bindings or the REST API.
- Agentic Workflow Integration: The model is designed to work within agentic workflows, providing deterministic handling for model outputs.
- Model Family Compatibility: Clef-flash is part of a model family that includes the larger Clef and multimodal Clef-omni models.
Use cases
- Developers use Clef-flash to build high-speed classification systems that categorize multimodal content like images and text into predefined labels.
- AI agent builders integrate Clef-flash to handle state transitions by turning environment data into structured decisions with associated probabilities.
- Security teams deploy Clef-flash to analyze unstructured log or event data and make real-time decisions about traffic or account activity.
- E-commerce platforms use Clef-flash to process product images and descriptions to automatically populate structured database fields based on a schema.
Specifications
- Modalities
- Text, JSON, Images, Video → Decisions with probabilities
- Parameters
- 9B
- Sizes and variants
- Clef (27B), Clef-flash (9B), Clef-omni (30B MoE)
- License
- Open-source decision models hosted on Workers AI
- Access
- Workers AI (Workers Bindings, REST API)
- Released
- 2026-10-01
- Architecture
- Multimodal decision model
- Input Modalities
- Text, JSON, Images, Video
- Output Type
- Probabilities for typed question schemas
- Platform
- Cloudflare Workers AI
- Inference Location
- Global edge network (330+ cities)
- Access Methods
- Workers Bindings, REST API
How it works
- Define Decision Schema: Define a schema of typed questions and the allowed options for the decisions the model needs to make.
- Provide State Input: Provide the current state of the application or environment as text, JSON, images, or video.
- Execute Inference: Call the model through Workers AI using the REST API or Workers Bindings to process the state against the schema.
- Receive Structured Decisions: Receive a structured response containing probabilities for every allowed option, enabling deterministic application logic.
FAQ
What is Clef-flash?
Clef-flash is a 9B-parameter multimodal decision model designed for high-speed classification and agentic workflows. It turns unstructured state data into structured decisions based on a schema of typed questions.
What modalities does Clef-flash support?
Clef-flash supports state inputs provided as text, JSON, images, or video. It processes these inputs to return probabilities for allowed options defined in a question schema.
How does Clef-flash compare to the standard Clef model?
Clef-flash is optimized for speed and cost-efficiency. Cloudflare recently lowered Clef-flash pricing and boosted inference speeds for the Clef family by up to 2.0x.
How can I access Clef-flash?
Developers can use Clef-flash through Cloudflare Workers AI. It is accessible via Workers Bindings for serverless applications or through a REST API for external service integration.
Added by AntSpace from public sources (developers.cloudflare.com).