Clef

Cloudflare's open-source decision models

Visit website

Cloudflare's open-source 27B multimodal decision model: state plus typed questions in, probabilities for every option out.

Clef model page on HuggingFace

Clef model page on HuggingFace

1 / 2

Overview

What is Clef?

Cloudflare's open-source decision models

Clef is Cloudflare's family of open-source decision models. Where chat models generate free-form text, a decision model takes a state plus a schema of typed questions and returns decisions: it reads the state as text, JSON, images or video, then returns a probability for every allowed option of every question in a single forward pass - no text generation, no output parsing.

The flagship Clef is a 27B multimodal model post-trained from Qwen3.8-27B (vision encoder included, standard sharded safetensors), released under Apache-2.0 with weights on HuggingFace. A smaller, faster variant called Clef-Flash is also available. The Clef API is fully compatible with Jev and SystemOne, and Cloudflare publishes a Decision Index leaderboard to compare decision models.

Alongside the models, Cloudflare announced a new RL fine-tuning platform so teams can extend Clef's capabilities with reinforcement learning on their own tasks. The announcement was published on the Cloudflare blog on October 1, 2026.

Platforms and languages

Clef Availability

Platforms

Api

Languages

English

Capabilities

Clef Key Features

Decisions, not text

Give it a state and a schema of typed questions; it returns a probability for every allowed option of every question in a single forward pass - no text generation, no output parsing.

Multimodal state input

Reads the state as text, JSON, images or video, so decisions can be grounded in whatever context you have.

Open weights, Apache-2.0

27B model post-trained from Qwen3.8-27B with vision encoder, released under Apache-2.0 on HuggingFace; Clef-Flash is the smaller, faster variant.

Jev and SystemOne API compatible

Drop-in compatibility with the Jev and SystemOne APIs for decision-model workloads.

RL fine-tuning platform

Cloudflare's new RL fine-tuning platform extends Clef's capabilities with reinforcement learning on your own tasks, plus a Decision Index leaderboard.

Best for

Who uses Clef?

ML engineers and developers building agentic systems

Get reliable, structured decisions with calibrated probabilities instead of parsing free-form model text

Plans and access

Clef Pricing

Free

Common questions

Clef FAQs

What is a decision model?

A model that turns a state and a schema of typed questions into decisions: it reads the state (text, JSON, images, video) and returns a probability for every allowed option of each question in a single forward pass - no free-form text generation and no output parsing.

What is the license?

Apache-2.0. The weights are open and published on HuggingFace (Cloudflare/clef).

How do I use Clef?

Get the weights from HuggingFace (Cloudflare/clef, 27B, BF16 safetensors) or use the Clef API, which is fully compatible with Jev and SystemOne. There is also a smaller, faster Clef-Flash variant.

What is the RL fine-tuning platform?

Cloudflare announced a new RL fine-tuning platform alongside Clef, designed to extend the models' capabilities with reinforcement learning on your own tasks. Details are in the Cloudflare blog announcement.

Reviews

See what the community thinks and share your experience.

—

Based on 0 ratings

Rating distribution

0
0
0
0
0

Leave a review

Sign in to rate

Community reviews

No reviews

No written reviews yet.

Explore by category

Llms Foundation Models

View all Llms Foundation Models websites