Nous: Hermes 4 70B

Nous: Hermes 4 70B, built on Meta-Llama-3.1-70B, demonstrates strong overall performance with a focus on hybrid reasoning. It performs among the fastest models, ranking in the 72nd percentile for speed, and offers highly cost-effective solutions, placing in the 76th percentile for price. The model exhibits exceptional reliability with a 95% success rate, indicating consistent and usable responses. In terms of specific benchmarks, Hermes 4 70B excels in acknowledging uncertainty, achieving 98.0% accuracy in Hallucinations (73rd percentile), and demonstrates robust Email Classification with 98.0% accuracy (68th percentile). Its Ethics performance is also very strong at 99.0% accuracy (58th percentile). Key strengths lie in its ability to avoid hallucination and its high reliability. However, the model shows notable weaknesses in complex domains such as Coding (55.0% accuracy, 21st percentile), Mathematics (69.0% accuracy, 27th percentile), and Reasoning (54.0% accuracy, 32nd percentile), despite its training emphasis on verified reasoning data. General Knowledge and Instruction Following also present areas for improvement, with accuracies around the 50th percentile. The model's support for JSON mode, schema adherence, and function calling, coupled with greater steerability, enhances its utility for structured tasks.