{
  "id": 11939003,
  "title": "Cloudflare unveils Clef models to rival Jev",
  "url": "https://urgent.news/2026/10/04/cloudflare-unveils-clef-models-to-rival-jev",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-04T12:17:50.000Z",
  "source": {
    "name": "Arabian Post",
    "slug": "arabian-post",
    "url": "https://thearabianpost.com/cloudflare-unveils-clef-models-to-rival-jev/"
  },
  "original_language": "en",
  "account": "Cloudflare has launched two open-source AI decision models, Clef and Clef-flash, aiming to compete with TypeSafe AI's Jev in the fast-growing market of structured decision-making in software and autonomous-agent workflows. Both models were released on October 1 and made available through Cloudflare's Workers AI platform. Clef has 27 billion parameters, while Clef-flash, designed for speed, has 9 billion. Both models accept a state and typed questions, returning probabilities for allowed answers instead of generating free-form text. They are compatible with Jev and System One API, allowing developers with existing Jev integrations to switch models by changing the endpoint and model identifier. The models can handle yes-or-no, multiple-choice, and ordered-score questions and can process up to 64 questions in a single request. Cloudflare claims an advantage in speed and benchmark performance, with median latencies of 209.3 milliseconds for Clef and 38.8 milliseconds for Clef-flash, compared to 524.1 milliseconds for Jev. Clef outperformed Jev on most of 10 decision benchmarks, recording higher scores on tests like BFCL case-exact and BANKING77 macro-F1. However, Jev still led in some tasks, such as When2Call and BRIGHT evaluations. The architecture of Clef and Clef-flash is designed for a narrower role, focusing on scoring allowed options in a single forward pass, which reduces latency and makes the models suitable for tasks like routing customer-support requests, identifying security incidents, or deciding whether an automated process should escalate to a human. The models are multimodal, capable of processing text, JSON, images, or video alongside a schema of questions. Cloudflare is offering a reinforcement-learning fine-tuning service to allow design partners to adapt Clef to their own workloads.",
  "summary": "Cloudflare has released two open-source artificial intelligence decision models, Clef and Clef-flash, mounting a direct challenge to TypeSafe AI’s Jev in the emerging market for fast, structured decisions inside software and autonomous-agent workflows. The company made both models available through Workers AI on October 1 and released their weights on Hugging Face under the Apache 2.0 licence.…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}