DeepHermes 3 Llama 8B Preview
DeepHermes 3 Llama 8B Preview combines normal chat with reasoning activated through a system prompt. Its weights remain published as a preview.
Preview
Training and model behavior
Nous describes the reasoning training as distillation from DeepSeek R1. The card covers annotation, judgment and function calls, and warns that the early reasoning behavior can have quirks.DeepHermes 3 Llama 8B Preview model card
Working with this model
Practical guidance. Use the required system prompt to compare reasoning with direct answers. Keep output budgets explicit; a long reasoning trace can consume the space needed for the final answer.
Download formats
Nous also publishes GGUF packages for this model. These are deployment formats of the same release.DeepHermes 3 Llama 8B Preview GGUF package
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Preview |
| Launch date | February 2025 |
| Access | Downloadable |
| Model identifiers | NousResearch/DeepHermes-3-Llama-3-8B-Preview |
| Context window | 131,072 tokens Configured position limit; practical serving limits depend on the runtime. DeepHermes 3 Llama 8B Preview configuration |
| Base model | meta-llama/Meta-Llama-3.1-8BDeepHermes 3 Llama 8B Preview model card |
| Repository license label | llama3DeepHermes 3 Llama 8B Preview model card |
Related models
- DeepHermes 3 Llama 3B Preview
Nous Research · DeepHermes
A preview model with direct-answer and extended-reasoning modes.
Launch date PreviewDownloadable - DeepHermes 3 Mistral 24B Preview
Nous Research · DeepHermes
A preview model with direct-answer and extended-reasoning modes.
Launch date PreviewDownloadable