Cloudflare announced Clef and Clef-flash on October 1: decision models that accept text and images, with a 64,000-token context, available on Workers AI with Apache 2.0 weights. The company also introduced reinforcement-learning fine-tuning for these decision tasks.
Its website-classification demonstration reports 2.2 seconds for a fetch-render-classify workflow against 4.7 seconds using gpt-oss-120b. Those are Cloudflare’s end-to-end demo timings. The practical attraction is a model built to assign decisions and probabilities inside a workflow; actual accuracy and probability calibration on a particular dataset still need measurement. Cloudflare’s model announcement and demo.