Cloudflare launches Clef, an open-source multimodal model built on Qwen

Cloudflare has launched Clef, an open-source multimodal decision model, including two versions: Clef and Clef-flash. Clef uses Qwen as its base model, with Clef built on Qwen3.8-27B and Clef-flash on Qwen3.5-9B.

In the Jev Decision Index benchmark, Clef takes the lead. Both models are fully compatible with Jev-API, making it easy for developers to test and verify. Clef comes with a native visual encoder that can directly parse image inputs and classify visual content, while the current Jev only supports text classification.

Clef expands the context window to 64k, compared to Jev's 32k, allowing developers to load richer system input states for more comprehensive judgments. The company says Clef delivers high accuracy and strong overall performance, matching or surpassing similar models in multiple benchmarks.

Cloudflare also released a new reinforcement learning product, letting customers fine-tune Clef for their specific business scenarios.