Nous: Hermes 3 405B Instruct

Nous: Hermes 3 405B Instruct, a frontier-level finetune of Llama-3.1 405B, demonstrates a strong overall performance profile. In terms of speed, it exhibits competitive response times, ranking in the 47th percentile across benchmarks. Its pricing is also competitive, placing in the 46th percentile. A standout feature is its exceptional reliability, achieving a perfect 100th percentile, indicating minimal technical failures and consistent usable responses. Across benchmark categories, Hermes 3 shows particular strength in classification and ethical reasoning, achieving perfect 100% accuracy in both Email Classification and Ethics. It also performs very well in General Knowledge (98.5%) and Coding (85.0%). While its Instruction Following (63.0%) and Reasoning (70.0%) scores are solid, they represent areas with some room for improvement compared to its top performances. Notably, in Email Classification and Ethics, it is highlighted as the most accurate model at its price point and among models of similar speed. Its advanced agentic capabilities, improved roleplaying, multi-turn conversation, and long context coherence are key differentiators, building on the Hermes 2 series with enhanced function calling and structured output.