IBM releases Granite 4.1 3B and 8B models with LLMON markup

A prompt-injection-resistant open model series could set a new baseline for agent security.

IBM released Granite 4.1 3B and 8B models built around LLMON, an LLM-native markup language designed to resist prompt injection by structuring model input. Weights are on Hugging Face.

Name

Granite 4.1 3B, Granite 4.1 8B

Params

3B, 8B

Domain

general

Read it at Hugging Face Blog →

Coverage: 2 sources, in order

1. 12 Nov 19:07 Hugging Face Blog Read the article →
2. 12 Nov 19:30 Unite.AI Read the article →

← Front page