Hermes-3-Llama-3.1-8B

NousResearch Llama 3 Community License out 2 years ago

8.03B parameters, text in, text out, 128K tokens of context. 1 provider serves it.

Verified 2026-10-25 Rank #205 in LLMs · score 88.5 method v1 Sources 3 Downloads 30d 464.1k · #170 of 619 tracked

Can I run it?4.8GB at INT4Estimatefits a 16 GB cardmethod

Can I use it?Llama 3 Community License'Built with Llama' attribution; >700M MAU clause; naming rule for derivativeslicence file2026-10-24

Do people use it?464.1kdownloads, 30 days#170 of 619 we track · top 27%+4 likes, downloads −7% this week522 likesHugging Face2026-10-24

Worth reading

Picked by rule from 11 tracked pieces: the lab's own post, the most shared, a field report, the latest.

Most discussed · 146 HN pts

Hermes 3: The First Fine-Tuned Llama 3.1 405B Model

Nous Research launched Hermes 3, the first full-parameter fine-tune of Meta's Llama 3.1 405B model, trained on an 8-node Lambda 1-Click Cluster.

lambdalabs.com 2024-08-15 146 HN pts · 67 comments 530 words about the Hermes 3 family family

Latest · 2 years ago

Nous Hermes 3 and exploiting underspecified evaluations

Nous Research released Hermes 3, fine-tuned Llama 3.1 models in 8B, 70B, and 405B sizes designed for role-playing and agentic function-calling.

Nous Hermes 3 training data consists primarily of synthetically generated responses.
The 8B model is likely the most popular version because better base models need less data for known tasks.

interconnects.ai 2024-08-16 2 HN pts 2.1k words about the Hermes 3 family family

11 best of 11 tracked

The 3 picks above, plus these 8. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

ArticleDeepHermes 3 - Llama-3.1 8B (non-reasoning)artificialanalysis.ai · undated
ArticleHermes 3 – Nous Researchnousresearch.com · 2024-08-15 · 29 HN pts · official
DiscussionHermes 3twitter.com · 2024-08-15 · 9 HN pts
ArticleAI Agents Explained: Where Hermes 3 Fits In 2026 | Markaicodemarkaicode.com · 2026-05-03
ArticleHermes 3 Technical Report — Paper Graphpaper-graph-ui.vercel.app · undated
Articlefinalend/hermes-3-llama-3.1:8b - Ollamaollama.com · undated
ArticleUnveiling Hermes 3: The First Fine-Tuned Llama 3.1 405B Model is on Lambda’s Cloudlearngrowthrive.net · 2024-08-16
ArticleHermes 3 Technical Report [Quick Review]liner.com · 2024-08-15

Show all 8

Where to run it

Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.

#	Cheapest live row	In /1M	Out /1M	Context	Quant	Seen
1	Featherless	—	—	—	—	Oct 25

All 1 provider →

Where it's used

Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.

Served by 1 API provider; heavily run locally (52 quantized builds); a big base for builders (370 fine-tunes, adapters and merges).

As a service1Featherless
On own hardware52quantized builds
Built on37039 fine-tunes · 282 adapters · 49 merges
In apps100+Hugging Face Spaces

Top Spaces running it: KingNish/OpenGPT-4o ♥ 4,302 · pliny-the-prompter/obliteratus ♥ 621 · Nymbo/Serverless-TextGen-Hub ♥ 44 · CZLC/BenCzechMark ♥ 26

Will it fit?

Weights only, an estimate: leave room for the KV cache. Method

Fits a 16 GB card at INT4

About 4.8 GB at INT4, 9.6 GB at FP8, 19 GB at BF16: weights only, so leave room for the KV cache.

Estimate params × bytes per param × 1.2 · how we estimate

The story so far

Weights, every dated article, and repo commits on one rail.

6 dated articles and 27 repo commits so far

Teal is the weights (2024-07-28), amber triangles are 6 dated articles (each links to it), grey ticks are repo commits.

What people built on it

422 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.

422 derivatives on the Hub, most of them adapters

From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.

GGUF bartowski/Hermes-3-Llama-3.1-8B-GGUF for llama.cpp, Ollama, LM Studio 102.5k downloads MLX mlx-community/Hermes-3-Llama-3.1-8B-MLX for Apple silicon 1.1k downloads AWQ solidrust/Hermes-3-Llama-3.1-8B-AWQ for vLLM, SGLang 334.6k downloads

All variants →

Reference

Everything else, folded: the full tables, every source.

Benchmarks

7 rows

Benchmarks 7 rows

measured third-party leaderboard rowauthor-reported number from the model card or eval files (7)No confidence intervals unless the source gave them; settings differ between rows, so compare within a benchmark, not across. Date = the benchmark's own date when it gives one, else when we read it.

Benchmark	Value	Setting / notes	Reported by	Date
author-reported Avg.	23.49	—	model author	2026-10-07
author-reported BBH (3-Shot)	30.72 percent	—	model author	2026-10-07
author-reported GPQA Diamond · Reason through hard problems	6.38 percent	—	model author	2026-10-07
author-reported IFEval · Chat and follow instructions	61.7 percent	—	model author	2026-10-07
author-reported MATH-500 · Reason through hard problems	4.76 percent	—	model author	2026-10-07
author-reported MMLU-Pro · Reason through hard problems	23.77 percent	—	model author	2026-10-07
author-reported MuSR (0-shot)	13.62 percent	—	model author	2026-10-07

What the card says

1 cited sentence

Summary generated, cited

The model uses the ChatML prompt format and is released under the Llama 3 Community License.^[1]^[2]

“Hermes 3 uses ChatML as the prompt format”
license

Generated 2026-10-07 15:04 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.

Use-case evidence

5 use cases

What it is good at 5 use cases

Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.

Facts

21 fields

Listed in LLMs

Facts

Hermes-3-Llama-3.1-8B has 8.03B parameters. (safetensors metadata, observed 2026-10-24)
Context length is 131,072 tokens (128K). (config.json max_position_embeddings, observed 2026-10-24)
License: Llama 3 Community License (friction: attribution). 'Built with Llama' attribution; >700M MAU clause; naming rule for derivatives. (license file, observed 2026-10-24)
Created 2024-07-28, last modified 2024-09-08. (HF API, observed 2026-10-24)

Pipeline tag	text-generation
Library	transformers
Architectures	LlamaForCausalLM
Model type	llama
Dtype	bfloat16
Params by dtype	BF16: 8,030,261,248
Size band	M
MoE	no
Context length	131,072 (config.json max_position_embeddings)
Languages	en
Papers	arXiv:2408.11857
License	Llama 3 Community License (llama3)
Hub snapshot	2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · LlamaForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · LlamaForCausalLM · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed

vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.

Hardware fit

3 estimates

Hardware fit ESTIMATE

Weights-only memory for Hermes-3-Llama-3.1-8B: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	19 GB · Single GPU (<24 GB)
FP8 (1 byte/param)	9.6 GB · Single GPU (<24 GB)
INT4 (0.5 bytes/param)	4.8 GB · Single GPU (<24 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.

All variants

40 variants

Lineage

Fine-tuned from meta-llama/Llama-3.1-8B Family: NousResearch/Hermes-3 422 derivatives on the Hub

Base model: meta-llama/Llama-3.1-8B

quantized: 52finetune: 39merge: 49adapter: 282

Top variants (40 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
solidrust/Hermes-3-Llama-3.1-8B-AWQ	quantized	AWQ	8.03B	M	—	solidrust	3	334.6k
mlc-ai/Hermes-3-Llama-3.1-8B-q4f16_1-MLC	quantized	—	—	—	—	mlc-ai	1	173.2k
bartowski/Hermes-3-Llama-3.1-8B-GGUF	quantized	GGUF	—	—	llama3	bartowski	22	102.5k
unsloth/Hermes-3-Llama-3.1-8B-bnb-4bit	quantized	-bnb	8.25B	M	llama3.1	unsloth	6	72.6k
mlabonne/Hermes-3-Llama-3.1-8B-lorablated	finetune	—	8.03B	M	—	mlabonne	44	58.4k
legraphista/Hermes-3-Llama-3.1-8B-IMat-GGUF	quantized	GGUF	—	—	llama3	legraphista	1	26.8k
mradermacher/Hermes-3-Llama-3.1-8B-i1-GGUF	quantized	GGUF	—	—	llama3	mradermacher	4	21k
mradermacher/Hermes-3-Llama-3.1-8B-GGUF	quantized	GGUF	—	—	llama3	mradermacher	6	16k
mlc-ai/Hermes-3-Llama-3.1-8B-q4f32_1-MLC	quantized	—	—	—	—	mlc-ai	0	14.5k
vicgalle/Configurable-Hermes-3-Llama-3.1-8B	finetune	—	8.03B	M	llama3.1	vicgalle	7	13.6k
QuantFactory/Hermes-3-Llama-3.1-8B-lorablated-GGUF	quantized	GGUF	—	—	—	QuantFactory	6	12.4k
Saxo/Linkbricks-Horizon-AI-Korean-Advanced-8B	quantized	GGUF	8.03B	M	apache-2.0	Saxo	0	12.4k
backyardai/Hermes-3-Llama-3.1-8B-GGUF	quantized	GGUF	—	—	llama3	backyardai	2	8.7k
squ11z1/Hypnos-i1-8B	quantized	GGUF	8.03B	M	apache-2.0	squ11z1	15	5.9k
mookiezii/Discord-Hermes-3-8B	finetune	—	8.03B	M	llama3	mookiezii	8	5.3k
QuantFactory/Aspire1.2-8B-TIES-GGUF	merge	GGUF	—	—	—	QuantFactory	1	5.2k
QuantFactory/Hermes-3-Llama-3.1-8B-Kor-Finance-Advisor-GGUF	quantized	GGUF	—	—	llama3.1	QuantFactory	1	4.7k
featherless-ai-quants/NousResearch-Hermes-3-Llama-3.1-8B-GGUF	quantized	GGUF	—	—	—	featherless-ai-quants	1	3.5k
informatiker/Hermes-3-Llama-3.1-8B-AWQ-INT4	quantized	AWQ	8.03B	M	llama3	informatiker	0	3.2k
Etherll/Herplete-LLM-Llama-3.1-8b	finetune	—	8.03B	M	—	Etherll	6	2.8k
justatom/Hermes-3-Llama-3.1-8B-Q8_0-GGUF	quantized	Q8_0	—	—	llama3	justatom	1	2.8k
lowlevelgoblin/Hermes-3-Llama-3.1-8B-Q4_K_M-GGUF	quantized	GGUF	—	—	llama3	lowlevelgoblin	1	2.8k
QuantFactory/Albatross2.1-8B-Instruct-GGUF	quantized	GGUF	—	—	—	QuantFactory	2	2.7k
tda45/TdAI	quantized	GGUF	8.03B	M	gpl-3.0	tda45	0	2.5k
LoganResearch/Nous-Hermes-ReflexAgent-8B-v1	adapter	—	8.03B	M	other	LoganResearch	6	2.2k
liodon-ai/Hermes-3-Llama-3.1-8B-imatrix-GGUF	quantized	imatrix	—	—	other	liodon-ai	0	2.4k
hlyn-labs/prompt-injection-judge-8b	quantized	GGUF	8.03B	M	llama3.1	hlyn-labs	2	2.2k
Saxo/Linkbricks-Horizon-AI-Nous-Hermes-3-Llama3.1-Korean-cpt-8b	finetune	—	8.03B	M	apache-2.0	Saxo	1	2.1k
itlwas/Hermes-3-Llama-3.1-8B-Q4_K_M-GGUF	quantized	GGUF	—	—	llama3	itlwas	1	1.9k
adamo1139/Hermes-3-Llama-3.1-8B-FP8-Dynamic	quantized	-FP8	8.03B	M	llama3.1	adamo1139	0	1.6k
Yingyaeliae/L3.1-Dark-Reasoning-LewdPlay-evo-Hermes-R1-Uncensored-8B-GGUF-heretic	quantized	GGUF	—	—	other	Yingyaeliae	1	1.5k
alfiannajih/g-retriever-resume-reviewer	finetune	—	8.06B	M	—	alfiannajih	2	1.5k
ClaudioItaly/Hermes-3-Llama-3.1-8B-Q5_K_M-GGUF	quantized	GGUF	—	—	llama3	ClaudioItaly	1	1.4k
Gunulhona/Llama-Ko-Merge	merge	—	8.03B	M	—	Gunulhona	1	1.3k
Mykee/HOTHUN-Hermes-3-8B-v1.1-GGUF	quantized	GGUF	—	—	llama3	Mykee	0	1.3k
Gunulhona/Hermes-Llama-Merge	merge	—	8.03B	M	—	Gunulhona	0	1.2k
mlx-community/Hermes-3-Llama-3.1-8B-MLX	finetune	MLX	8.03B	M	llama3	mlx-community	1	1.1k
aedmark/vsl-cryosomatic-hypervisor	quantized	GGUF	—	—	unlicense	aedmark	1	1.1k
Rybens/Hermes-3-Llama-3.1-8B-Q3_K_S-GGUF	quantized	GGUF	—	—	llama3	Rybens	1	1.1k
AbteeXAILab/lumynax-chat-hermes-3-llama31-8b-gguf	quantized	gguf	—	—	llama3.1	AbteeXAILab	0	1.1k

Repository changelog

27 commits

Repository changelog 27 commits

Date	Commit	SHA
2024-09-08	adding closing curly bracket for tool call json	896ea44
2024-09-08	fix typo with missing closing curly bracket	4898392
2024-09-08	prevent double-stringification of tool_call arguments in jinja template	ebcf7c1
2024-09-08	fixing issue with extra curly bracket and removing requirement for tool call name with tool role	5ba4922
2024-09-05	Adding Evaluation Results (#12)	cc68dc2
2024-09-05	fixing chat template system role issue	5e3fd48
2024-08-23	Update README.md	aabb745
2024-08-23	Update README.md	610bec1
2024-08-22	Update tokenizer_config.json	aa50cfc
2024-08-19	Update tokenizer_config.json	67bf4ac
2024-08-18	Update tokenizer_config.json	f537e10
2024-08-15	Update README.md	8c61760
2024-08-15	Update README.md	387a2e7
2024-08-14	Update README.md	01d6dcd
2024-08-07	Update README.md	7f5471a
2024-08-07	Update README.md	1a34dfa
2024-08-06	Update tokenizer_config.json	db6a44d
2024-08-02	Update config.json	248bb9b
2024-08-01	Update README.md	cd22519
2024-07-31	Update tokenizer_config.json	9949f48
2024-07-31	Update tokenizer_config.json	5aada9f
2024-07-31	Update config.json	34ccd90
2024-07-31	Update generation_config.json	d4ed47f
2024-07-31	Update special_tokens_map.json	39cc98e
2024-07-28	Create README.md	06f644a
2024-07-28	Upload folder using huggingface_hub	46ed7f4
2024-07-28	initial commit	d9ecc06

All providers

1 provider · 1 endpoint

Where to run 1 providers · 1 endpoints

Hosted endpoints observed for Hermes-3-Llama-3.1-8B, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting.

Provider	Model ref	Context	Quant	$/M in	$/M out	tok/s	TTFT	Tools	Status	Observed
Featherless hf	NousResearch/Hermes-3-Llama-3.1-8B	—	—	—	—	—	—	—	live	2026-10-25

Articles and coverage

11 pieces

Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.

Official1

Hermes 3 – Nous Research

nousresearch.com2024-08-15officialabout the Hermes 3 familyHN 29 pts · 3 comments

Analysis8

Hermes 3: The First Fine-Tuned Llama 3.1 405B Model

lambdalabs.com2024-08-15HN front pageabout the Hermes 3 familyHN 146 pts · 67 comments530 words

Nous Research launched Hermes 3, the first full-parameter fine-tune of Meta's Llama 3.1 405B model, trained on an 8-node Lambda 1-Click Cluster. The model achieves a 50% reduction in VRAM and disk requirements through Neural Magic's FP8 quantization, enabling single-node operation. Hermes 3, which has been downloaded over 33 million times, is available for free via Lambda's Chat Completions API and Lambda Chat interface.written, reviewed, cited

DeepHermes 3 - Llama-3.1 8B (non-reasoning)

artificialanalysis.aiauthority759 words

DeepHermes 3 - Llama-3.1 8B Preview (Non-reasoning) is an open weights model released on February 13, 2025, by Nous Research that scores 5 on the Artificial Analysis Intelligence Index. It features a 128k tokens context window with knowledge up to December 2023 and operates under the LLAMA 3.1 COMMUNITY LICENSE AGREEMENT. The model supports text input and output but ranks at the lower end compared to comparable non-reasoning models with a median score of 6.written, reviewed, cited

The model classifies as Small with 8B parameters, falling within the 4B–40B parameters range for open weights comparisons.
Pricing tiers for proprietary comparisons use a blended 3:1 input/output price ratio to determine cost categories.
The model explicitly does not support image input and is not multimodal, processing only text.

Nous Hermes 3 and exploiting underspecified evaluations

interconnects.ai2024-08-16authorityabout the Hermes 3 familyHN 2 pts · 0 comments2,069 words

Nous Research released Hermes 3, fine-tuned Llama 3.1 models in 8B, 70B, and 405B sizes designed for role-playing and agentic function-calling. The models utilize a training mix where 69% of tokens are output tokens, yet internal evaluations show the 8B model trails Llama 3.1 Instruct on standard chat templates. This release highlights the difficulty of evaluating frontier models due to discrepancies between reported benchmarks and independent testing.written, reviewed, cited

Nous Hermes 3 training data consists primarily of synthetically generated responses.
The 8B model is likely the most popular version because better base models need less data for known tasks.
Using LoRA adapters for DPO training is expected to cause performance degradation relative to full fine-tuning.

AI Agents Explained: Where Hermes 3 Fits In 2026 | Markaicode

markaicode.com2026-05-03about the Hermes 3 family2,239 words

Hermes 3 Technical Report — Paper Graph

paper-graph-ui.vercel.appabout the Hermes 3 family

Hermes 3 Llama 3.1 8B — local inference guide | RunLocalAI

runlocalai.co562 words

finalend/hermes-3-llama-3.1:8b - Ollama

ollama.comabout the Hermes 3 family430 words

Hermes 3 Technical Report [Quick Review]

liner.com2024-08-15about the Hermes 3 family814 words

Community1

Hermes 3

twitter.com2024-08-15about the Hermes 3 familyHN 9 pts · 0 comments

All coverage tracked for this model (11) — append-only, nothing is ever removed

Title	Source	Published	First seen	Signals
DeepHermes 3 - Llama-3.1 8B (non-reasoning)	artificialanalysis.ai	—	2026-10-07	exa
Hermes 3 Technical Report — Paper Graph	paper-graph-ui.vercel.app	—	2026-10-07	exa
Hermes 3 Llama 3.1 8B — local inference guide \| RunLocalAI	runlocalai.co	—	2026-10-07	exa
finalend/hermes-3-llama-3.1:8b - Ollama	ollama.com	—	2026-10-07	exa
AI Agents Explained: Where Hermes 3 Fits In 2026 \| Markaicode	markaicode.com	2026-05-03	2026-10-07	exa
Nous Hermes 3 and exploiting underspecified evaluations	interconnects.ai	2024-08-16	2026-10-07	hn
Unveiling Hermes 3: The First Fine-Tuned Llama 3.1 405B Model is on Lambda’s Cloud	learngrowthrive.net	2024-08-16	2026-10-07	exa
Hermes 3: The First Fine-Tuned Llama 3.1 405B Model	lambdalabs.com	2024-08-15	2026-10-07	hn
Hermes 3	twitter.com	2024-08-15	2026-10-07	hn
Hermes 3 – Nous Research	nousresearch.com official	2024-08-15	2026-10-07	hn
Hermes 3 Technical Report [Quick Review]	liner.com	2024-08-15	2026-10-07	exa

Papers

1 paper

Papers 1

arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).

Hermes 3 Technical Report this model

Ryan Teknium, Jeffrey Quesnelle, Chen Guang · 2024-08-15 · PDF · HF paper page (60 upvotes)

Similar models

3 models

Similar models

Llama-3.1-8B-Instruct · 8.03B
same size band (M), same task (text-generation), more providers
Qwen3-8B · 8.19B
same size band (M), same task (text-generation), more providers, permissive license
Meta-Llama-3-8B-Instruct · 8.03B
same size band (M), same task (text-generation), more providers

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:36 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:04 UTC · Last verified 2026-10-25

Follow Hermes-3-Llama-3.1-8B

Watch this model

/feeds/models/NousResearch--Hermes-3-Llama-3.1-8B.xml(RSS 2.0) · every change to Hermes-3-Llama-3.1-8B
/feeds/families/NousResearch--Hermes-3.xml(RSS 2.0) · every change across the Hermes-3 family

Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.