Hermes 3 - Llama-3.1 8B Overview

Hermes 3 is the latest iteration in Nous Research's flagship Hermes series, a generalist language model based on the Llama-3.1 architecture with 8 billion parameters and a 32768 token context length. This version introduces substantial improvements over Hermes 2, focusing on advanced agentic capabilities, enhanced roleplaying, superior reasoning, and more coherent multi-turn conversations. A core philosophy of the Hermes series is user alignment, providing powerful steering and control to the end-user.

Key Capabilities

• Advanced Agentic Capabilities: Designed for complex task execution and autonomous interaction.
• Improved Roleplaying & Reasoning: Offers more nuanced and consistent character interactions and logical problem-solving.
• Enhanced Multi-turn Conversation & Long Context Coherence: Maintains context and relevance over extended dialogues.
• Reliable Function Calling: Supports robust function calling with a structured prompt format, enabling integration with external tools.
• Structured Output (JSON Mode): Capable of generating responses strictly adhering to a provided JSON schema.
• ChatML Prompt Format: Utilizes ChatML for structured multi-turn chat, compatible with OpenAI API formats.

Benchmarks

Hermes 3 demonstrates competitive performance against Llama-3.1 Instruct models across general capabilities, with specific strengths in various areas. For detailed results, refer to the Hermes 3 Technical Report.

Good For

• Applications requiring advanced agentic behavior and tool use.
• Complex conversational AI and roleplaying scenarios.
• Generating structured data outputs in JSON format.
• Developers familiar with OpenAI's ChatML API format.