Hermes 3 Llama 3.2 3B is a compact, instruction-tuned model with a strong emphasis on following complex directions and structured outputs. NousResearch's fine-tuning gives it a noticeably sharper conversational focus than the base Llama model, handling multi-step reasoning and roleplay scenarios with more consistency. Its small size means it runs efficiently on modest hardware, though it trades raw capability for accessibility.
| Benchmark | Score | Type | Recorded |
|---|---|---|---|
| IFEval | 38.2 | accuracy | 6d ago |