Hermes 2 Pro Llama 3 8B

Hermes 2 Pro is an 8 billion parameter model built on Meta's Llama 3, designed for developers building tool-using and agentic systems. The model reaches 90% on Nous Research's function calling evaluation, built in partnership with Fireworks.AI, and 84% on its structured JSON output evaluation, enabling reliable schema-constrained generation. Training combined an updated and cleaned OpenHermes 2.5 dataset with a newly introduced function-calling and JSON-mode dataset, using DPO and RLHF techniques. The model uses the ChatML prompt format and adds dedicated agentic tokens (<tools>, <tool_call>, <tool_response>), making it well suited to structured, multi-turn agent workflows. With an 8,192 token context window and general assistant capabilities alongside tool use, Hermes 2 Pro fills the niche between smaller models lacking agentic features and larger models, offering a practical balance for production systems.