{
  "id": 19694251,
  "title": "LLMON: An LLM-Native Markup Language, now in Granite 4.1 3B and 8B",
  "url": "https://urgent.news/2026/11/12/llmon-an-llm-native-markup-language-now-in-granite-4-1-3b-and-8b",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-12T19:07:35.000Z",
  "source": {
    "name": "Hugging Face",
    "slug": "hugging-face",
    "url": "https://huggingface.co/blog/ibm-granite/llmon-an-llm-native-markup-language-now-in-granite"
  },
  "original_language": "en",
  "account": "LLMON, an LLM-native markup language, has been developed to address the brittleness and unexpected failures of LLMs when processing complex context. This markup language uses meaningful tags and reserved special-token delimiters to mark semantically coherent components, such as data schemas, nesting, unique IDs, and instructions. By encoding structure in separate reserved tokens rather than characters, LLMON allows models to process nested structures more accurately and efficiently. Two experimental preview models, granite-4.1-3b-LLMON-preview and granite-4.1-8b-LLMON-preview, have been released, trained to consume and generate LLMON-marked text. These models come with their own chat template, enabling users to build and render LLMON-marked messages as usual. LLMON's instruction/data separation is designed to minimize indirect prompt injection attacks, as the model can now clearly distinguish between instructions and data payloads. Benchmarks, such as OPI, ASB, and StruQ, show that LLMON-previews significantly reduce vulnerability to such attacks compared to other models. The performance of LLMON models remains on par with Granite 4.1 Instruct models across standard benchmark categories.",
  "summary": null,
  "key_points": [
    "LLMON is an LLM-native markup language to improve processing complex context",
    "Granite 4.1 3B and 8B models released with LLMON preview versions",
    "LLMON reduces vulnerability to indirect prompt injection attacks"
  ],
  "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."
}