Cloudflare debuts open-weight multimodal decision models Clef and Clef-flash, claiming they are smarter and faster than Jev, based on Qwen3.8-27B and Qwen3.5-9B
Sure, it costs more, but it can handle images and video and it's available on Hugging Face if you have the hardware horsepower to run it locally
Context & Ripple Effects
Cloudflare had already joined Vercel and others in adding Jev after it made AI tool selection faster and cheaper. Its new models turn that earlier integration into a direct product challenge, rather than leaving Cloudflare dependent on a third-party decision model.
The release builds on a Qwen3.8-27B base noted for tool use, vision and coding and the broader availability of open-weight multimodal Qwen models. Cloudflare is pairing that model lineage with Workers AI hosting and local Hugging Face availability.
First-order effects
- Cloudflare customers gain Clef and Clef-flash as alternatives to Jev for classification and agentic workflows, including tasks that use images or video.
- Jev faces a directly competing decision-model offering from a platform that had only weeks earlier been among the providers adding Jev.
Second-order effects
- Vercel and other Jev adopters will have to compare Clef's claimed speed and intelligence against Jev's workflow performance, especially where multimodal inputs matter.
- Organizations with sufficient local hardware can evaluate a self-run open-weight option alongside hosted decision-model services, increasing buyer leverage over deployment and model choice.
Third-order effects
- If Cloudflare's performance claims hold in production, decision models may become less defensible as standalone products and more differentiated by the hosting, deployment and workflow layers around open-weight bases.
- The combination of an open model, a managed Workers AI route and local availability points toward a split market: reusable model weights underneath proprietary operating and distribution layers.
The trend: AI decision workflows are shifting toward open-weight models wrapped in proprietary inference, deployment and control-plane services.