# Hugging Face — LLMON: An LLM-Native Markup Language, now in Granite 4.1 3B and 8B

- Company: Hugging Face (huggingface.co)
- Announced: 2026-11-12T19:07:35+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://huggingface.co/blog/ibm-granite/llmon-an-llm-native-markup-language-now-in-granite
- Record: https://forck.live/items/35483-llmon-an-llm-native-markup-language-now-in-granite-4-1-3b-and-8b
- Subject: Platform
- Models affected: granite-4.1-3b-LLMON-preview, granite-4.1-8b-LLMON-preview, Granite 4.1 Instruct

IBM released two experimental preview models derived from Granite 4.1 base models, trained to consume and generate LLMON, an LLM-native markup language that marks context components with meaningful tags. The company says LLMON exists because context structure is usually implicit or encoded in formats such as JSON that were not built to be digested by LLMs, which it says makes smaller models brittle and open to prompt injection. IBM says indirect prompt injections hidden in user data are mostly ignored by the LLMON models because instructions and data are explicitly separated, and that on some benchmarks attack success rates drop to nearly zero while regular instruction-tuned models get hijacked with high probability. IBM says this holds while general capability is maintained relative to Granite 4.1 Instruct models. LLMON does not replace chat templates but enriches them: the preview models ship with their own chat template, so a turn can name its parts and mark each one exec: True or False, with True parts bound into an exec block stating which instruction runs and what it operates on. IBM also released a GitHub repo with a toolkit, workflows and templates, plus converters and notebooks that translate JSON to LLMON and back losslessly, and Mellea has native LLMON support enabled by automatically LLMONizing its LLM calls.

## Evidence

Verbatim from https://huggingface.co/blog/ibm-granite/llmon-an-llm-native-markup-language-now-in-granite:

> LLMs rely on their context to generate the next set of tokens. Over a long trajectory, the context fills with numerous types of components: a system prompt, past conversation turns, instructions that act on retrieved documents, tool responses or user data, requirements constraining the allowed output, and reasoning chains. The context carries a rich structure, most of which is either implicit to the model or encoded in formats such as JSON that were not designed to be digested by LLMs. This is one of the main reasons LLMs, particularly those on the smaller end of the scale, are brittle and sometimes fail in unexpected ways. The lack of explicit structure can also be exploited by attackers, for example, in the form of prompt injections.
> 
> To address this, we developed LLMON (LLM Object Notation, pronounced “lemon”), an LLM-native markup language for explicitly communicating structure and semantics to the models. Semantically coherent components are marked with meaningful tags which are wrapped in reserved special-token delimiters. Data schemas, including nesting, are stated in a way that is unambiguous to a causal LLM.

---

Record: https://forck.live/items/35483-llmon-an-llm-native-markup-language-now-in-granite-4-1-3b-and-8b
Catalogue: https://forck.live/llms.txt
Current issue: https://forck.live/feed.md
