Hermes 4 70B
Hermes 4 70B mixes reasoning and direct-response data. Its post-training supports tool calls, JSON output and instructions that control response style.
Training and model behavior
The Hermes 4 corpus contains about five million examples and sixty billion tokens. The model can place deliberation inside think tags. Its chat template can enable reasoning and preserve earlier reasoning segments. Tool calls use explicit tags for parsing.Hermes 4 70B model card
Working with this model
Practical guidance. Keep reasoning settings and chat formatting fixed when comparing this release with its base. Check tool names and arguments before execution. Higher willingness to answer is a behavior choice, not a correctness measure.
Download formats
Nous also publishes FP8 packages for this model. These are deployment formats of the same release.Hermes 4 70B FP8 package
Specifications & access
| Detail | Value and conditions |
|---|---|
| Launch date | |
| Access | Downloadable |
| Model identifiers | NousResearch/Hermes-4-70B |
| Context window | 131,072 tokens Configured position limit; practical serving limits depend on the runtime. Hermes 4 70B configuration |
| Base model | meta-llama/Meta-Llama-3.1-70BHermes 4 70B model card |
| Repository license label | llama3Hermes 4 70B model card |
Related models
- Hermes 4 405B
Nous Research · Hermes 4
A Llama 3.1 405B derivative with optional reasoning and structured output.
Launch date Downloadable - Hermes 4.3 36B
Nous Research · Hermes 4
A Seed-OSS derivative post-trained across the Psyche network.
Launch date Downloadable - Hermes 4 14B
Nous Research · Hermes 4
A Qwen3-14B derivative with optional reasoning and structured output.
Launch date September 2025Downloadable