Nous Research is the most prominent community-driven open-weight research lab in 2026, with the Hermes lineage as its flagship finetune family and a growing portfolio of original architecture and infrastructure work. The 2026 Nous lineage includes Hermes 3 (Llama and Qwen backbones), Hermes 4, DeepHermes, the Forge Reasoning API, Psyche distributed training, and several specialised finetune families. This page consolidates the lineage and the position.
Key Findings
- Hermes 3 is the most-downloaded community finetune family on Hugging Face, with Hermes 3 Llama 405B as the largest and Hermes 3 8B / 70B / 3B covering common deployment sizes.
- Hermes 4 (released late 2025) integrates reasoning modes plus enhanced tool use, making it the leading community-derived alternative to first-party reasoning models.
- DeepHermes (released early 2026) combines Hermes finetune recipe with reasoning training; available in 3B, 8B, and 24B sizes covering on-device through cloud deployment.
- The Forge Reasoning API is Nous Research\u2019s hosted inference offering, layering ensemble reasoning techniques on top of Hermes-finetuned models for production deployment.
- Psyche is Nous Research\u2019s distributed pretraining framework, enabling community-pooled training compute across heterogeneous hardware via the Solana blockchain coordination layer.
Nous Research Hermes Family (May 2026)
| Model | Base | Capability | License |
|---|---|---|---|
| Hermes 4 405B | Llama 3.1 405B | Frontier-tier finetune | Llama 3.1 Community |
| Hermes 4 70B | Llama 3.1 70B | General-purpose finetune | Llama 3.1 Community |
| Hermes 4 14B | Qwen2.5 14B | Mid-size finetune | Apache 2.0 (Qwen) |
| Hermes 4 7B | Qwen2.5 7B | Small finetune | Apache 2.0 (Qwen) |
| Hermes 3 405B | Llama 3.1 405B | Frontier finetune | Llama 3.1 Community |
| Hermes 3 70B | Llama 3.1 70B | General-purpose | Llama 3.1 Community |
| Hermes 3 8B | Llama 3.1 8B | Small finetune | Llama 3.1 Community |
| Hermes 3 3B | Llama 3.2 3B | Tiny finetune | Llama 3.2 Community |
| DeepHermes 24B | Mistral Small 3 | Reasoning finetune | Apache 2.0 (Mistral) |
| DeepHermes 8B | Llama 3.1 8B | Reasoning finetune | Llama 3.1 Community |
| DeepHermes 3B | Llama 3.2 3B | Reasoning finetune small | Llama 3.2 Community |
| Genstruct | Various | Synthetic data generation | Various |
Hermes Recipe
The Hermes finetune recipe emphasises strong system-prompt steerability, role-play coherence, function calling, and structured output. Practical differences from base models: more reliable JSON output, better adherence to long custom system prompts, more diverse stylistic range. The recipe is documented and reproducible; Nous Research publishes datasets and code with most releases.
Forge Reasoning API and Psyche
| Product | Status |
|---|---|
| Forge Reasoning API | Hosted inference with ensemble reasoning techniques (MCTS, beam search, voting) on Hermes models |
| Psyche distributed pretraining | Open-source framework for community-pooled pretraining across heterogeneous compute |
| Nous DisTrO optimiser | Distributed training optimiser reducing inter-node bandwidth requirements ~857x |
| Hermes Vision (research) | Vision-language finetune family on Llama and Qwen VLM bases |
Strategic Context
Three patterns shape Nous Research\u2019s 2026 position. First, the community brand is durable: Nous Research is the highest-recognition non-corporate open-weight lab and benefits from sustained Discord and X engagement. Second, the technical work is increasingly original: DisTrO optimiser and Psyche framework move Nous from a finetune lab to a distributed-training infrastructure lab. Third, the monetisation through Forge Reasoning API plus community subscription models is testing whether community-driven AI labs can sustain commercial revenue without becoming closed.
Brand Visibility Implications
Nous Research is a highly-cited brand in technical AI communities. AI assistant queries about "best open-source AI finetunes", "Hermes model", "community AI training", and similar terms drive sustained traffic from developer audiences. Brands selling AI training infrastructure, finetuning services, AI community platforms, and decentralised AI face strong AI-mediated discovery surface for this category.
Methodology
Model data compiled from primary Hugging Face model card disclosures and the Nous Research GitHub and Discord disclosures through 23 May 2026. Updated quarterly.
How Presenc AI Helps
Presenc AI monitors brand visibility on Nous Research and community AI queries across ChatGPT, Claude, Gemini, and Perplexity. For AI training infrastructure brands, finetuning service vendors, and decentralised AI projects, the platform identifies the prompts driving research-traffic patterns and the gaps where new content unlocks share of voice.