Cloudflare Releases Open Source Multimodal Model Clef
Cloudflare has launched Clef, an open-source multimodal decision-making model based on Qwen. The model is designed to process both text and images, enhancing analytical capabilities.
Cloudflare, a company focused on web infrastructure and security, has released Clef, a new open-source multimodal decision-making model. This model is built upon the Qwen architecture and is available in two versions: Clef and Clef-flash.
The Clef models are engineered with native visual encoders, allowing them to directly interpret image inputs and classify visual content, a departure from previous text-only models. Furthermore, Clef features an expanded context window of 64,000 tokens, enabling more comprehensive analysis and decision-making compared to earlier models with 32,000 tokens.
In benchmarking tests using the Jev Decision Index, Clef models have demonstrated leading performance. Both versions are fully compatible with the Jev-API, facilitating easier testing and integration for developers. Cloudflare also introduced a new reinforcement learning product, enabling customers to fine-tune Clef models for their specific business needs.
The models utilize Qwen as their base, with Clef based on Qwen3.8-27B and Clef-flash on Qwen3.5-9B. Cloudflare states that Clef offers high precision and robust performance, rivaling or surpassing similar models in various benchmarks. Details on the models and their performance metrics are available on Hugging Face.