Clef is a 27B multimodal decision model from Cloudflare, post-trained from Qwen/Qwen3.8-27B and released under Apache-2.0. Instead of generating text, it takes a state (text, JSON, images or video) and a schema of typed questions and returns a probability for every allowed option of every question in a single forward pass, with no free-form generation and no output parsing. Questions are of three types: noul (true/false), choice (named options) and score (ordered options). Architecturally it keeps the Qwen3.8-27B backbone with its vision encoder and adds a small joint schema head that reads the final hidden states, routes evidence from the state to each question and scores all options jointly; a softmax per question gives probabilities. The README says its API is fully compatible with Jev and SystemOne, and a systemone helper accepts a POST /v1/systemone request body and returns answers keyed by question ID. Cloudflare's announcement says Clef has a vision encoder where Jev does text classification only today, and a 64k context window compared with Jev's 32k; encode_record defaults to a 16,384-token max_length. In the README's internal run of the Decision Index 0.2.1 suite, Clef scores 98.5 on BFCL, 94.2 macro-F1 on BANKING77 (Jev 79.7) and 97.4 on CLINC150+OOS (Jev 89.3), with median latency of 209.3 ms against 524.1 ms for Jev. It trails Jev on several reasoning-style tasks, for example GPQA Diamond (48.0 vs 78.3), MMLU-Pro (65.9 vs 82.7) and BBH (73.7 vs 92.9). In Cloudflare's own domain-classification workflow for its Threat Intelligence team, Clef took 2.2s to fetch, render and classify a site where gpt-oss-120b took 4.7s. A smaller, faster variant, Clef-Flash, is also published, and Cloudflare hosts both on Workers AI. The usage example was tested with torch 2.11 and transformers 5.10.2 on a single H200, loading the model through joint_schema_model.py with load_release_model.