- Who
- Hugging Face
- What
- released two experimental Granite 4.1 LLMON preview models, in 3B and 8B sizes
What happened
IBM is releasing two experimental preview models, granite-4.1-3b-LLMON-preview and granite-4.1-8b-LLMON-preview, trained to consume and generate text marked with LLMON, an LLM-native format that separates instructions from data. On the OPI benchmark, the post reports that the previews avoided 47% more attacks than the best tested frontier model and 95% more than the best tested open-source model. The post says the models are for exploration and evaluation, not production, and that adaptive attacks remain under development.
Why it matters
The post describes how LLMON separates instructions from data and reports its preview models’ results on static injection benchmarks, while noting that adaptive attacks remain an active area of work.
Read next
See the full storyHugging Face Releases Two Experimental Granite 4.1 LLMON Preview ModelsThe story so far
Hugging Face released experimental Granite 4.1 preview models in 3B and 8B sizes, trained to read and generate text marked with LLMON, a format that separates instructions from data. The models are available for exploration and evaluation, but Hugging Face says they are not ready for production. Hugging Face reported that on the OPI benchmark, the previews avoided 47% more attacks than the best tested frontier model and 95% more than the best tested open-source model. The company said adaptive attacks remain under development.