Nous: Hermes 4 405B

Nous: Hermes 4 405B, a large-scale reasoning model built on Meta-Llama-3.1-405B, demonstrates competitive performance across various benchmarks. It exhibits competitive response times, ranking in the 59th percentile for speed, and offers moderate pricing, placing it in the 34th percentile. A standout feature is its exceptional reliability, achieving a 100% success rate across all benchmarks, indicating consistent and usable responses. The model excels in specific areas, achieving perfect accuracy in Ethics (Baseline) and near-perfect scores in Email Classification (99.0%) and General Knowledge (99.5%). These results highlight its strong capabilities in ethical reasoning, classification tasks, and broad factual recall. Its expanded post-training corpus, emphasizing reasoning traces, appears to contribute to its strong performance in these domains. While demonstrating solid performance in Coding (83.0% accuracy), its Instruction Following (61.0%) and Reasoning (60.0%) scores, though respectable, suggest areas for potential improvement compared to its top-tier performance in other categories. The introduction of a hybrid reasoning mode with internal deliberation offers flexibility, and its support for structured outputs like JSON mode and function calling enhances its utility for developers.