Hermes 4
Overview
Hermes 4 is an open-source family of large language models developed by Nous Research, released in August 2025. Built upon Meta's Llama 3.1 architecture, Hermes 4 offers 70-billion and 405-billion parameter variants, both featuring a context window of 131,072 tokens. A distinctive feature of Hermes 4 is its hybrid reasoning mode, which allows the model to toggle between standard responses and explicit reasoning using `<think>...</think>` tags, enabling deeper deliberation on complex tasks. This design aims to enhance performance in areas such as mathematics, coding, and general reasoning. The model is available through Nous Research's platform, with pricing set at $0.05 per million input tokens and $0.20 per million output tokens for the 70-billion parameter variant, and $0.09 per million input tokens and $0.37 per million output tokens for the 405-billion parameter variant. As of June 2026, Hermes 4 is recognized among leading open-weight reasoning models, demonstrating competitive performance in various benchmarks. Its open-source nature and advanced reasoning capabilities position it as a notable option for applications requiring extensive context processing and complex problem-solving. In comparison to its peers, Hermes 4's hybrid reasoning mode and large context window provide advantages in tasks necessitating detailed analysis and extended input sequences. However, the absence of specific benchmark scores and tool-use support may limit its applicability in certain structured workflows. Nonetheless, its open-source availability and innovative design make it a valuable resource for developers and researchers seeking advanced language model capabilities.