On October 1 (according to local time), Cloudflare introduced Clef, an open-source multimodal decision-making model grounded in Qwen. This model comes in two iterations: Clef and Clef-flash. Clef is constructed upon the foundation of Qwen3.8-27B, whereas Clef-flash leverages Qwen3.5-9B. In the benchmark assessment of the Jev Decision Index 0.2.1, Clef emerged as the frontrunner. Notably, both models are entirely compatible with the Jev-API, streamlining the process for developers to conduct tests and validations. Clef inherently incorporates a visual encoder, allowing it to directly interpret image inputs. Moreover, its context window has been expanded to 64k, bolstering its overall decision-making prowess.
