Nous Research’s newest product is a fine-tuned version of Meta’s Llama 3.1-8B model, designed to enhance long-chain reasoning, structured function calling, and user-controlled reasoning depth. It is the first model with a toggle-on reasoning feature, allowing seamless switching between fast, intuitive responses and detailed, step-by-step reasoning—all within a single LLM.
In with the New, (not) Out with the Old
DeepHermes-3 Preview builds upon Hermes 3’s function-calling foundation. Hermes 3 was fine-tuned from Llama 3.1-8B and designed as an instruct model, making it effective at interacting with external tools and APIs. DeepHermes-3 builds on this by introducing notable improvements in reasoning, structured decision-making, and response quality:
Toggle-able Reasoning Model
-
The new creation introduces a dedicated system prompt that activates deep thinking mode. This enables the model to engage in extended chains of thought before answering.
-
This feature is not present in Hermes 3, which primarily functioned as an instruct-tuned model without deep-thinking ability.
-
Uses <think></think> tags to document its internal monologue, showing e
