{
  "id": 12816444,
  "title": "Cloudflare Open Sources Decision Models for AI Agents",
  "url": "https://urgent.news/2026/10/08/cloudflare-open-sources-decision-models-for-ai-agents",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-08T06:54:00.000Z",
  "source": {
    "name": "InfoQ",
    "slug": "infoq",
    "url": "https://www.infoq.com/news/2026/10/clef-decision-models/"
  },
  "original_language": "en",
  "account": "Cloudflare has released Clef, a collection of open-weight AI models intended to select from predetermined choices rather than generate text. The company unveiled both 9B and 27B parameter models, accompanied by a platform enabling customization of the models for distinct decision-making tasks. Clef functions as a 27B multimodal model, accepting a state and a structured set of questions as input and providing a decision with associated probabilities for each permitted option in response. This decision model categorizes inputs and outputs typed results with corresponding probabilities, empowering AI agents to utilize these outcomes to determine appropriate actions, such as routing a support request, escalating it, or deferring the decision to a human. Cloudflare's API is interoperable with Typesafe AI's Jev System One model.\n\nMichelle Chen, a group product manager at Cloudflare, alongside Alex Reneau, a principal machine learning engineer, and Kevin Flansburg, a senior engineering manager, highlighted that since Clef is hosted on Cloudflare's infrastructure, they benefit from their GPUs at the edge, resulting in low network latency and swifter decision-making. Consequently, developers can incorporate Clef into the active path of decision-making agents and pair it with one of Cloudflare's LLMs on Workers AI to execute actions. The company also unveiled Clef-Flash, a more compact 9B multimodal model tailored for latency-sensitive decisions, achieving a median latency of 38.8 ms in Cloudflare's assessments, as opposed to the 209.3 ms latency for the 27B Clef model.\n\nClef distinguishes itself from other decision models by incorporating a vision encoder, allowing it to process images and classify visual content, a capability absent in earlier models like Jev, which solely performs text classification. Additionally, Clef offers a larger 64K context window compared to Jev's 32K, facilitating the inclusion of more input state for the model to classify. Community feedback on a popular Hacker News thread questioned whether this constitutes a novel model category, with some arguing that it is merely an underutilized aspect that Typesafe AI has popularized. However, others lauded the model's potential, emphasizing the importance of evaluating its performance in real-world scenarios, particularly in identifying whether the model appropriately recognizes when to defer decisions to a human. Cloudflare intends to fine-tune Clef for particular use cases, such as support triage and bot classification, utilizing its labeled historical data to enhance accuracy and speed.",
  "summary": "During its recent “Birthday Week”, Cloudflare announced Clef, a set of open-weight AI models designed to choose between predefined options rather than generate text. Cloudflare released 9B- and 27B-parameter models, along with a platform for adapting them to specific decision-making tasks. By Renato Losio",
  "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."
}