opus-research/opus-moderation-4-neo
Text Classification • 0.4B • Updated • 15
Opus Moderation 4
Content moderation + jailbreak detection, plain from_pretrained. om4-large moderates, om4-neo catches attacks - run them in parallel.
Note Jailbreak specialist - 0.938 jailbreak F1, beats our 1.7B flagship at 395M
opus-research/opus-moderation-4-large
Text Classification • 0.4B • Updated • 14 • 1Note Recommended general moderator - 0.543 macro F1, beats every public baseline tested
opus-research/opus-moderation-4-fast
Text Classification • 0.1B • Updated • 27 • 1Note 149M CPU variant - the smallest unified toxicity+jailbreak model we know of