{"slug": "cloudflare-releases-clef-and-clef-flash-open-weight-decision-models-that-return", "title": "Cloudflare Releases Clef and Clef-flash: Open-Weight Decision Models That Return Typed Probabilities Instead of Text", "summary": "Cloudflare released Clef and Clef-flash, its first models trained by the Workers AI team, as open-weight decision models under Apache 2.0 that return typed probabilities instead of free-form text. Clef is post-trained from Qwen3.8-27B and Clef-flash from Qwen3.5-9B, both supporting noul, choice and score question types with up to 64 questions and 4 images per Workers AI request, and both are compatible with TypeSafe AI's Jev API. On Cloudflare's 10-benchmark Decision Index 0.2.1 shortlist a Clef model scored highest on 7, including BANKING77 macro-F1 of 94.20 versus Jev's 79.74, though Jev leads GPQA Diamond 78.3 to 48.0; all numbers are vendor-reported with no independent replication yet.", "body_md": "Cloudflare has released [Clef and Clef-flash](https://blog.cloudflare.com/clef-decision-models/), the first models trained by its Workers AI team. They are decision models, not chatbots. Each reads an input state and a schema of typed questions. It returns a probability for every allowed answer, with no free-form text. Both are open-weight under Apache 2.0 and compatible with TypeSafe AI’s Jev API.\n\n**Is it deployable?** Yes, Both models run today on [Workers AI](https://developers.cloudflare.com/workers-ai/models/clef/), and the weights are on [Hugging Face](https://huggingface.co/Cloudflare/clef) for self-hosting.\n\n## **What a Decision Model Does**\n\nAn LLM generates tokens one at a time, and its output still needs parsing. A decision model only answers a fixed set of questions about an input. **Clef supports 3 question types:**\n\n- `noul` : yes/no, returns the probability of yes.\n- `choice` : picks 1 named option, with per-option probabilities and a confidence value.\n- `score` : rates against an ordered rubric, returning a probability-weighted score.\n\nOn Workers AI, 1 request carries up to 64 questions and up to 4 images.\n\nTypeSafe AI launched [Jev](https://typesafe.ai/blog/introducing-system-one-models-and-jev), its first ‘System One’ model, on September 15, 2026. Open alternatives like [Kev-9B](https://huggingface.co/jaredpalmer/kev-9b) and [Laya](https://huggingface.co/convaiinnovations/laya) followed. Clef uses the same System One API. Switching from Jev means changing the endpoint and model name.\n\n## **How Clef Works**\n\nClef is post-trained from [Qwen3.8-27B](https://huggingface.co/Qwen/Qwen3.8-27B), and [Clef-flash](https://huggingface.co/Cloudflare/clef-flash) from [Qwen3.5-9B](https://huggingface.co/Qwen/Qwen3.5-9B). Both keep the backbone’s vision encoder.\n\nInference has 2 stages. The backbone first runs a single prefill-only pass over the state and questions. A small transformer, the joint schema head, then reads the final hidden states. It routes evidence to each question, lets fields cross-attend, and scores all options jointly. A per-question softmax turns logits into probabilities.\n\nTraining froze both backbones and jointly optimized the routing head with rank-256 low-rank adapters. The loss pairs label-smoothed cross-entropy with a Brier loss for calibration. A secondary objective, Reinforcement Learning for Calibrated Decisions (RLCD), gives partial credit to adjacent ordinal choices.\n\n## **Interactive Explainer**\n\n## **Benchmarks: Where Clef Wins and Where It Does Not**\n\nOn Cloudflare’s 10-benchmark shortlist from the [Decision Index](https://clef-evals.workers-ai-mle.workers.dev/) 0.2.1 suite, a Clef model scored highest on 7.\n\n- **BANKING77 (macro-F1):** Clef 94.20 vs Jev 79.74.\n- **CLINC150+OOS (macro-F1):** Clef 97.43 vs Jev 89.27.\n- **Home appliances (case exact):** Clef-flash 97.73 vs Jev 52.27.\n\nJev keeps clear leads elsewhere. The [full model card](https://huggingface.co/Cloudflare/clef) shows Jev ahead on GPQA Diamond (78.3 vs 48.0). It also leads MMLU-Pro (82.7 vs 65.9) and BBH (92.9 vs 73.7).\n\nOn TypeSafe’s own [workflow evals](https://evals.typesafe.ai/), Clef beat Jev in 3 of 4 areas, by small margins. Invoice processing was 64.7 vs 61.8, customer service 76.3 vs 76.0, and security incidents 62.9 vs 61.7. Jev leads agent trace observability, 71.6 vs 68.5.\n\nIn Cloudflare’s threat intelligence workflow, Clef classified a domain in 2.2 seconds. gpt-oss-120b took 4.7 seconds.\n\n*All numbers are vendor-reported, with no independent replication yet.*\n\n## **Feature Comparison**\n\n| Feature | Clef | Clef-flash | Jev | Kev-9B | Laya | \n|---|---|---|---|---|---|\n| **Developer** | Cloudflare | Cloudflare | TypeSafe AI | Jared Palmer | Convai Innovations | \n| **Size** | 27B | 9B | Not disclosed | 9B + 45.4M LoRA | 421M | \n| **Backbone** | Qwen3.8-27B | Qwen3.5-9B | Not disclosed | Qwen3.5-9B-Base | ModernBERT-large | \n| **Weights** | Apache 2.0 | Apache 2.0 | Hosted API | Apache 2.0 | Apache 2.0 | \n| **Image input** | Yes | Yes | No | No | No | \n| **Context** | 65,536 | 65,536 | 32K (per Cloudflare) | 65,536 (8,192 validated) | 512 (English) | \n| **Median latency*** | 209.3 ms | 38.8 ms | 524.1 ms | 51.4 ms | 5.8 ms | \n| **Hosted price (input)** | $0.24/M | $0.09/M | $0.042/M | Self-host | Self-host | \n\n**Cloudflare’s internal Decision Index run. All 5 implement the System One API. Sources: [Clef docs](https://developers.cloudflare.com/workers-ai/models/clef/), [Clef-flash docs](https://developers.cloudflare.com/workers-ai/models/clef-flash/), [Clef card](https://huggingface.co/Cloudflare/clef), [Jev post](https://typesafe.ai/blog/introducing-system-one-models-and-jev), [Kev-9B card](https://huggingface.co/jaredpalmer/kev-9b), [Laya card](https://huggingface.co/convaiinnovations/laya).*\n\n## **Deployment and Fine-Tuning**\n\nBoth models are callable through the Workers AI binding (`env.AI.run()`), the REST API, or [AI Gateway](https://developers.cloudflare.com/ai-gateway/). For self-hosting, the model cards list testing on a single H200 with BF16 weights.\n\nCloudflare also announced a reinforcement learning service for tuning Clef on private data. It starts with Cloudflare’s forward-deployed engineers, with a self-serve platform later. The pipeline combines AI Gateway, Workers AI, Containers and a new Trainer component. Teams can apply via the [design partner form](https://www.cloudflare.com/resource/clef-rl-interest).\n\n## **Key Takeaways**\n\n- Clef (27B) and Clef-flash (9B) are Apache 2.0, Jev-compatible decision models.\n- Median latency: 209.3 ms for Clef, 38.8 ms for Clef-flash, 524.1 ms for Jev.\n- Clef reads text, JSON, images and video within a 64K-token context window.\n- Jev still leads on knowledge-heavy tests like GPQA Diamond, MMLU-Pro and BBH.\n- An RL fine-tuning service starts with Cloudflare’s forward-deployed engineers.\n\nCheck out the **[Model weight](https://huggingface.co/Cloudflare/clef), [Demo](https://clef-evals.workers-ai-mle.workers.dev/)** and **[Technical details](https://blog.cloudflare.com/clef-decision-models/)**. All credit goes to the researcher of this project. Also, feel free to follow us on **[Twitter](https://x.com/intent/follow?screen_name=marktechpost)** and don’t forget to join our **[150k+ML SubReddit](https://www.reddit.com/r/machinelearningnews/)** and Subscribe to **[our Newsletter](https://magic.beehiiv.com/v1/f5e63dd4-5653-4f09-83e2-321a8b1ba526?email={{email}})**. Wait! are you on telegram? [now you can join us on telegram as well.](https://t.me/machinelearningresearchnews)\n\nNeed to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? [Connect with us](https://forms.gle/MJjjVDPS7whH8Ngs6)\n\nAsif Razzaq is the CEO of Marktechpost AI Media Inc.. As a visionary entrepreneur and engineer, Asif is committed to harnessing the potential of Artificial Intelligence for social good. His most recent endeavor is the launch of an Artificial Intelligence Media Platform, Marktechpost, which stands out for its in-depth coverage of machine learning and deep learning news that is both technically sound and easily understandable by a wide audience. The platform boasts of over 2 million monthly views, illustrating its popularity among audiences.", "url": "https://wpnews.pro/news/cloudflare-releases-clef-and-clef-flash-open-weight-decision-models-that-return", "canonical_source": "https://www.marktechpost.com/2026/10/01/cloudflare-releases-clef-and-clef-flash/", "published_at": "2026-10-02 01:00:41+00:00", "updated_at": "2026-10-02 01:14:14.551110+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "large-language-models", "ai-products", "ai-infrastructure"], "entities": ["Cloudflare", "Clef", "Clef-flash", "Workers AI", "TypeSafe AI", "Jev", "Qwen3.8-27B", "Qwen3.5-9B"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/cloudflare-releases-clef-and-clef-flash-open-weight-decision-models-that-return", "markdown": "https://wpnews.pro/news/cloudflare-releases-clef-and-clef-flash-open-weight-decision-models-that-return.md", "text": "https://wpnews.pro/news/cloudflare-releases-clef-and-clef-flash-open-weight-decision-models-that-return.txt", "jsonld": "https://wpnews.pro/news/cloudflare-releases-clef-and-clef-flash-open-weight-decision-models-that-return.jsonld"}}