Model components / Fine-tuned / chat models

Hermes-4-Llama-3.1-405B

Frontier hybrid-mode reasoning model built on Meta-Llama-3.1-405B. It is the largest Hermes 4 variant in the public release set, with 1.1K cumulative downloads and 269 likes on Hugging Face.

Hermes 4, hybrid-mode reasoning instruct model post-trained on Meta-Llama-3.1-405B (406B params). Released Sep 2 2025; single-checkpoint reason/non-reason toggle. Tech report arXiv:2508.18255. Verified live on HF June 2026. License = Llama3 (Llama 3.1 Community License, non-OSI, use-restricted). Consolidated on 2026-07-29 from hermes-4-14b, hermes-4-3-36b, hermes-4-llama-3-1-70b; openness follows hermes-4-llama-3-1-405b, the release that currently governs. See sources/slug_aliases.yaml.

Openness

3 high confidence

3.0

weights: open(Apache-2.0)
base: Seed-OSS-36B-Base
post-training-data: closed(proprietary ~5M-sample corpus)
code: partial(inference only)
license: Apache-2.0(OSI)

Nous ships one Hermes 4 post-training recipe on four independent base models, and the base determines the license the weights carry: Seed-OSS-36B and Qwen3-14B are Apache-2.0, while the Llama-3.1 405B and 70B builds inherit Meta's Llama-3.1-Community terms (700M-MAU commercial cap, acceptable-use policy, naming requirements). Scored 3 on the Apache builds rather than 2 on the Llama ones, because most-restrictive-across-SKUs is a rule about variants you cannot substitute away from - Qwen 2.5's restricted 72B caps Qwen because the 7B is not a substitute for the flagship - and here the base is a free choice. Anyone wanting Apache-licensed Hermes 4 takes the 36B or 14B build and gets it, so the family genuinely offers open weights. Not 4 or 5: the post-training corpus is proprietary and only inference code ships, so the recipe is not reproducible. Corrected from 2/restricted on 2026-07-29 after the tier consolidation initially applied most-restrictive across all four bases.

https://huggingface.co/NousResearch/Hermes-4-405B-FP8 recorded 2026-06-04
License: Llama3, base Meta-Llama/Llama-3.1-405B

Adoption

2 high confidence

2.0

~35,066 HF downloads/mo, the most-downloaded Hermes 4 SKU (small size + permissive license drive uptake). Early-adopter scale.

https://huggingface.co/NousResearch/Hermes-4-14B recorded 2026-06-04
35,066 downloads last month

Capability

4 medium confidence

4.0

Strongest Hermes 4 SKU; near-frontier on math/knowledge for an open instruct model, though coding (LiveCodeBench 61.3) trails the 2026 frontier coders. Not a category-defining 5 (those are DeepSeek-V4/Qwen3.6/Kimi-class). Vendor/tech-report figures, corroborated by OpenRouter.

https://arxiv.org/pdf/2508.18255 recorded 2026-06-04
Hermes 4 technical report with comprehensive benchmark evaluation across math/code/knowledge
https://openrouter.ai/nousresearch/hermes-4-405b recorded 2026-06-04
Hermes 4 405B benchmark figures (AIME, MMLU-Pro, LiveCodeBench)

Unchanged since 2026-07-29 (last edited, not re-checked)