← sources

X3Guard: A 150M parameter prompt guard

source: reddit (reddit:LocalLLaMA)published 2026-11-15 15:46ingested 2026-11-16 02:38

retained text

At Axiomic Labs we have just released X3Guard-150M, a prompt guard model trained on our high-performing base model, GPT-X3.

For any given prompt, X3Guard flags whether it is Harmful, and whether it is Adversarial, in one go. 

It outperforms models that are many times its size, and can run entirely on edge devices, within the browser, with response times in the tens of milliseconds on iPhone. 

[The model card is here](https://huggingface.co/AxiomicLabs/X3Guard-150M)

[You can use an in-browser demo here](https://huggingface.co/spaces/AxiomicLabs/X3Guard-150M-Demo) 

document fetched 2026-11-16 00:37 · canonical https://reddit.com/r/LocalLLaMA/comments/1xwyp7w/x3guard_a_150m_parameter_prompt_guard