Clef
27B decision model on a Qwen3.8-27B backbone with a vision encoder. Open alternative to TypeSafe's hosted Jev; self-hosting needs ~85 GB VRAM at 64k context (The Register). Benchmarks are Cloudflare's own.
- Context window
- 65,536 tokens
- Input
- text, image, video
- Output
- text
- License
- apache-2.0
- Pricing
- input: $0.24 (per 1M input tokens (Workers AI); output price not stated in the docs page checked) source
- Verified
- 2026-10-02
How to call it
| Provider | Model id | Endpoint / URL | Docs |
|---|---|---|---|
| Cloudflare Workers AI | @cf/cloudflare/clef | — | docs |
| Hugging Face | — | huggingface.co/Cloudflare/clef | — |
Notable capabilities (1)
- Multimodal decision model (typed option scoring): Takes text, JSON, images or video plus a schema of typed questions and returns a probability for every allowed option in one prefill-only forward pass; Jev-API compatible; Cloudflare reports 209 ms median latency vs Jev's 524 ms on its 43-benchmark set. source
Open-weights "decision model" from Cloudflare: it scores the supplied options instead of generating free text. Use it on Workers AI as @cf/cloudflare/clef or download the weights from Hugging Face (Apache 2.0). A smaller, faster sibling is cloudflare-clef-flash.
Timeline entry
- Cloudflare releases Clef and Clef-flash, open-weight multimodal decision models that are Jev-API compatible ★★★
On Oct 1, 2026 Cloudflare released Clef (built on Qwen3.8-27B) and Clef-flash (built on Qwen3.5-9B), Apache-2.0 open-weight "decision models". Like TypeSafe's Jev they return probabilities for typed options instead of generating text, but they also accept images and video and have a 64k context…