We present Hermes 4, a family of hybrid reasoning models that combinestructured, multi-turn reasoning with broad instruction-following ability. Wedescribe the challenges encountered during data curation, synthesis, training,and evaluation, and outline the solutions employed to address these challengesat scale. We comprehensively evaluate across mathematical reasoning, coding,knowledge, comprehension, and alignment benchmarks, and we report bothquantitative performance and qualitative behavioral analysis. To support openresearch, all model weights are published publicly athttps://huggingface.co/collections/NousResearch/hermes-4-collection-68a731bfd452e20816725728