← Back to journal

Nous Research releases Hermes 4

Hermes 4 combined open-weight reasoning with general instruction following across several model sizes.

Nous Research released Hermes 4 as a hybrid family that could reason at length or answer directly. The project published model weights and documented its data, training, and evaluation choices in a technical report.

The useful idea here is not that every prompt needs visible reasoning. It is that an open model can decide when extra thought helps and when a short answer is enough.