Join waitlist
Back to products
DeepHermes 3

DeepHermes 3

Intuitive responses and deep reasoning, in one model

Website huggingface.co
Overview

What it is

DeepHermes 3 from Nous Research is a Llama-3.1 8B based LLM with a toggleable reasoning mode for complex tasks. Combines fast responses with deep, chain-of-thought reasoning.

Intent

I need it when

Build AI applications with advanced reasoning and chain-of-thought capabilities

DeepHermes 3 unifies reasoning mode (long chains of thought via system prompt) and intuitive response mode in a single 8B model, enabling developers to toggle between deep analytical thinking and fast responses without model switching.

Implement function calling and structured JSON output in conversational AI

DeepHermes 3 includes specialized training for function calling and JSON mode with documented prompt formats, enabling reliable tool integration and structured data extraction in multi-turn conversations.

Integrate an LLM into existing Python ML pipelines with minimal dependencies

DeepHermes 3 integrates seamlessly with Hugging Face Transformers library and supports multiple inference frameworks (vLLM, SGLang, TGI), allowing drop-in deployment into existing PyTorch and ML workflows.

Deploy a lightweight, open-source LLM locally or on custom infrastructure

As a free, open-source 8B Llama-3 model available on Hugging Face, DeepHermes 3 can be deployed via Transformers, vLLM, SGLang, or Docker with full control over infrastructure and no vendor lock-in.

Improve LLM accuracy on complex reasoning tasks like math and logic problems

DeepHermes 3 Preview includes reasoning capabilities distilled from R1, with benchmarks showing significant gains on MATH and reasoning tasks when deep thinking mode is enabled via system prompt.

Drop

Not a fit when

  • User requires a fully managed, no-code API without infrastructure setup
  • User needs guaranteed production SLA and enterprise support without custom arrangement
  • User requires models larger than 8B parameters for their specific task
  • User cannot allocate GPU resources or lacks technical expertise to deploy locally
  • User needs real-time model updates or automatic version management
Commercials

Pricing

Free open-source model; optional paid API via Nous Research View pricing