Updated

Cloudflare Clef and Clef-flash

Cloudflare released Clef and Clef-flash, open-weight decision models hosted on Workers AI that return probabilities for typed choices instead of generating text. Clef adds image input and a 64K context window; the weights are available under Apache 2.0.

By No Code MBA · Facts checked

Provider
Cloudflare
Announcement date
Status at last check
Generally available
Model family
Clef
Capabilities
Text, Image

What changed

Clef (27B parameters) and Clef-flash (9B) are Cloudflare's first decision models. They read a state and up to 64 typed questions (yes/no, choice or score) and return probabilities for each allowed answer instead of generating text. Both have a 64K token context window, and Clef accepts up to four images through a vision encoder. Cloudflare reports median latency of 209.3 ms for Clef and 38.8 ms for Clef-flash, versus 524.1 ms for Jev.

Access and availability

Run them on Workers AI as @cf/cloudflare/clef and @cf/cloudflare/clef-flash via the env.AI.run() binding or the REST /ai/run endpoint, with AI Gateway support. Weights are on Hugging Face (Cloudflare/clef and Cloudflare/clef-flash) under Apache 2.0.

Official access documentation

Pricing

Clef: $0.24 per million input tokens, with no output token charge

Official pricing

What you can build

Evaluate one bounded task such as support-ticket routing against your current classifier or model, measuring decision quality, confidence calibration, latency, and how often a human should review the result.

Limitations

Clef-flash trades some precision for speed, and Cloudflare does not say whether it supports images. Images are capped at four per request and 1,024 tokens each, remote image URLs are not accepted, and requests whose media and questions exceed the context fail.

Sources and release history