Nous ResearchHermesQwen3ForCausalLM

Hermes 4 14B — Hardware Requirements & GPU Compatibility

ChatReasoningRoleplayFunctions

Hermes 4 14B is a 14.8B-parameter open language model from Nous Research in the Hermes family. It supports a context window of up to 40,960 tokens. At Q4_K_M it needs about 9.50 GB of VRAM — see which GPUs and Macs can run it below.

14.8B params · dense41K contextApache 2.0Aug 20253.0K downloads191 likes76.3K quant downloads

Based on Qwen3 14B

Your hardware

Verdicts below are for the reference RTX 4090. Set yours to personalise the page.

Detection reads your browser's GPU name with no data sent anywhere.

Runs great on a 24 GB GPU

Runs from 8 GB (IQ2_M · 16K)Best on 24 GB (Q4_K_M · ~69 tok/s · 40K)Macs from 16 GB (Q4_K_M · ~15 tok/s)

Apache 2.0

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run Hermes 4 14B

One recommended way. Every other method, OS and quant is in the step-by-step guide.

Other ways

each opens step-by-step

Run in cloud

Fits on Tesla V100 PCIe 16GB (4 GB headroom) · Q4_K_M

Generation speed: ~62 tok/s; generation speed
Cost per 1M output tokens: $0.56; per 1M output tokens

Deployon Vast.aiBest optionfrom $0.12/hr

Compare GPUs →

Run Hermes 4 14B

11.9 GB VRAM at 16K context · fits the RTX 4090 (12.1 GB free)

Works on macOS, Windows and Linux.

Install Ollama (Linux)
$curl -fsSL https://ollama.com/install.sh | sh
Install Ollama (macOS / Windows)
Download and run the app; it installs the ollama command.
Download Ollama ↗
Run Q4_K_M
$ollama run hf.co/bartowski/NousResearch_Hermes-4-14B-GGUF:Q4_K_M
Raise the context window
Type it inside the chat. For the API server instead, start it with OLLAMA_CONTEXT_LENGTH=16384 ollama serve.
$/set parameter num_ctx 16384
A larger context needs more memory than the weights alone.
Use the API
Ollama serves its API on http://localhost:11434.

Works on macOS, Windows and Linux.

Install LM Studio
Desktop app for macOS (Apple silicon), Windows and Linux.
lmstudio.ai ↗
Download the model
In the app press Ctrl/Cmd + Shift + M and paste https://huggingface.co/bartowski/NousResearch_Hermes-4-14B-GGUF, then pick Q4_K_M. Or use the CLI (run LM Studio once first):
$lms get https://huggingface.co/bartowski/NousResearch_Hermes-4-14B-GGUF@Q4_K_M
Open on Hugging Face ↗
Load and chat
Open the model loader, select the model and set Context Length to 16384. Leave GPU offload at maximum if it fits.
Start the local server (optional)
$lms server start

Works on macOS, Windows and Linux.

Install llama.cpp (macOS)
MacPorts, Nix and conda-forge also work.
$brew install llama.cpp
Install options ↗
Install llama.cpp (Windows)
$winget install llama.cpp
Install llama.cpp (Linux)
Download the prebuilt llama-*-bin-ubuntu-* archive for your GPU backend (CPU, Vulkan) from the releases page, unpack it and run the binaries from that folder. Homebrew (brew install llama.cpp) and conda-forge also work on Linux.
llama.cpp releases ↗
Start the server
Downloads the model on first run. The chat UI and an OpenAI-compatible API are at http://localhost:8080.
$llama-server -hf bartowski/NousResearch_Hermes-4-14B-GGUF:Q4_K_M -c 16384 -ngl 99 --port 8080

Works on macOS, Windows and Linux.

Option A: Docker Model Runner
Docker Desktop: Settings > AI > Enable Docker Model Runner. Docker Engine on Linux: install the docker-model-plugin package.
$docker model run hf.co/bartowski/NousResearch_Hermes-4-14B-GGUF:Q4_K_M
Model Runner docs ↗
The quant tag after the repo is how Hugging Face cards write it; if your Docker version rejects it, drop the tag and pick the quant by repo.
Option B: llama.cpp server image (NVIDIA)
Needs the NVIDIA Container Toolkit. Models are cached in the llama-cache volume. Open http://localhost:8080.
$docker run --gpus all -p 8080:8080 -v llama-cache:/root/.cache ghcr.io/ggml-org/llama.cpp:server-cuda -hf bartowski/NousResearch_Hermes-4-14B-GGUF:Q4_K_M -c 16384 -ngl 99 --host 0.0.0.0 --port 8080
Image tags (cuda13, vulkan, rocm) ↗
Cache path inside the image is not in the docs; remove the -v flag if the volume complains.

Linux only.

Install vLLM
Linux only; on Windows run it inside WSL2. NVIDIA GPU shown; AMD ROCm, Intel and CPU builds have their own install pages.
$uv venv --python 3.12 --seed
$source .venv/bin/activate
$uv pip install vllm --torch-backend=auto
vLLM installation ↗
Serve the model
OpenAI-compatible API at http://localhost:8000/v1.
$vllm serve NousResearch/Hermes-4-14B --max-model-len 16384
Runs the original weights, not the Q4_K_M GGUF; needs ~32 GB of GPU memory.

macOS only.

MLX supports only macOS, so follow these steps for that setup.

Install mlx-lm
$pip install mlx-lm
Chat in the terminal
$mlx_lm.chat --model mlx-community/Hermes-4-14B-8bit
Downloads the model on first run. The MLX repo has its own quantisation; it may not match the quant selected above.
Or start a local server
Listens on http://localhost:8080.
$mlx_lm.server --model mlx-community/Hermes-4-14B-8bit

Works on macOS, Windows and Linux.

Install Jan
jan.ai ↗
Find the model
Open the Hub tab, paste this model ID in Search, and choose Q4_K_M if offered.
$bartowski/NousResearch_Hermes-4-14B-GGUF
Use the local API (optional)
Settings > Local API Server > Start Server. It listens on http://127.0.0.1:1337.

Rent GPUs hourly to run Q4_K_M right there. Pricing comes from partner marketplaces and varies.

Available GPU rentals for Q4_K_M, prioritized by value
GPU	Speed	$/1M tokens	Rent
Tesla V100 PCIe 16GBCheapestCheapest4 GB free	~62 tok/s	$0.56	Deploy on Vast.aifrom $0.12/hr (opens in new tab)	4 GB free · ~62 tok/s · $0.56 / 1M tokens
RTX 5060 Ti 16GB4 GB free	~31 tok/s	$1.47	Deploy on Vast.aifrom $0.16/hr (opens in new tab)	4 GB free · ~31 tok/s · $1.47 / 1M tokens
RTX 4060 Ti 16GB4 GB free	~20 tok/s	$2.55	Deploy on Vast.aifrom $0.18/hr (opens in new tab)	4 GB free · ~20 tok/s · $2.55 / 1M tokens
RTX 4000 Ada (20 GB)8 GB free	~25 tok/s	$2.26	Deploy on RunPodfrom $0.20/hr (opens in new tab)	20 GB · 8 GB free · ~25 tok/s · $2.26 / 1M tokens
RTX 5070 Ti (16 GB)4 GB free	~61 tok/s	$0.92	Deploy on Vast.aifrom $0.20/hr (opens in new tab)	16 GB · 4 GB free · ~61 tok/s · $0.92 / 1M tokens
RTX 3090 (24 GB)12 GB free	~64 tok/s	$0.95	Deploy on RunPodfrom $0.22/hr (opens in new tab)	24 GB · 12 GB free · ~64 tok/s · $0.95 / 1M tokens
RTX A6000 (48 GB)36 GB free	~53 tok/s	$1.75	Deploy on RunPodfrom $0.33/hr (opens in new tab)	48 GB · 36 GB free · ~53 tok/s · $1.75 / 1M tokens
RTX 4090 (24 GB)12 GB free	~69 tok/s	$1.37	Deploy on RunPodfrom $0.34/hr (opens in new tab)	24 GB · 12 GB free · ~69 tok/s · $1.37 / 1M tokens

Compare GPUs →

How Much VRAM Does Hermes 4 14B Need?

Select a quantization level; the listed hardware adheres to it.

VRAM	+32K	+128K	File	Quality	On RTX 4090 (reference)
5.1 GB	10.1 GB	—	4.7 GB	2-bit · i-quant extra small — Importance-weighted 2-bit, extra small	Fits: 40K context · ~129 tok/s
5.3 GB	10.3 GB	—	5 GB	2-bit · i-quant small — Importance-weighted 2-bit, small	Fits: 40K context · ~125 tok/s
5.6 GB	10.7 GB	—	5.3 GB	2-bit · i-quant medium — Importance-weighted 2-bit, medium	Fits: 40K context · ~117 tok/s
6.4 GB	11.4 GB	—	5.9 GB	3-bit · i-quant — Importance-weighted 3-bit	Fits: 40K context · ~103 tok/s
6.7 GB	11.8 GB	—	6.4 GB	3-bit · i-quant extra small — Importance-weighted 3-bit, extra small	Fits: 40K context · ~97 tok/s
6.9 GB	11.9 GB	—	5.8 GB	2-bit · K-quant — 2-bit quantization with K-quant improvements	Fits: 40K context · ~95 tok/s
7.1 GB	12.1 GB	—	6.7 GB	3-bit small — 3-bit small quantization	Fits: 40K context · ~92 tok/s
7.3 GB	12.3 GB	—	6.9 GB	3-bit · i-quant medium — Importance-weighted 3-bit, medium	Fits: 40K context · ~90 tok/s
7.8 GB	12.9 GB	—	7.3 GB	3-bit medium — 3-bit medium quantization	Fits: 40K context · ~84 tok/s
8.0 GB	13.1 GB	—	8.5 GB	4-bit · legacy — 4-bit legacy quantization	Fits: 40K context · ~82 tok/s
8.2 GB	13.2 GB	—	7.9 GB	3-bit large — 3-bit large quantization	Fits: 40K context · ~80 tok/s
8.6 GB	13.6 GB	—	8.1 GB	4-bit · i-quant compact — Importance-weighted 4-bit, compact	Fits: 40K context · ~77 tok/s
8.9 GB	14 GB	—	9.4 GB	4-bit · legacy+ — 4-bit legacy quantization with offset	Fits: 40K context · ~73 tok/s
8.9 GB	14 GB	—	8.6 GB	4-bit small — 4-bit small quantization	Fits: 40K context · ~73 tok/s
8.9 GB	14 GB	—	8.5 GB	4-bit · i-quant non-linear — Importance-weighted 4-bit, non-linear	Fits: 40K context · ~73 tok/s
9.5 GB	14.5 GB	—	9 GB	4-bit medium — 4-bit medium quantization — most popular sweet spot	Fits: 40K context · ~69 tok/s
9.7 GB	14.7 GB	—	9.6 GB	4-bit large — 4-bit large quantization	Fits: 40K context · ~68 tok/s
10.8 GB	15.8 GB	—	10.3 GB	5-bit small — 5-bit small quantization	Fits: 40K context · ~61 tok/s
11.2 GB	16.2 GB	—	10.5 GB	5-bit medium — 5-bit medium quantization — good quality/size tradeoff	Fits: 40K context · ~59 tok/s
11.3 GB	16.4 GB	—	11 GB	5-bit large — 5-bit large quantization	Fits: 40K context · ~58 tok/s
12.8 GB	17.9 GB	—	12.1 GB	6-bit, very good — 6-bit quantization, very good quality	Fits: 40K context · ~51 tok/s
15.4 GB	20.4 GB	—	15.7 GB	8-bit, near-lossless — 8-bit quantization, near-lossless	Fits: 40K context · ~43 tok/s
30.2 GB	35.2 GB	—	29.5 GB	16-bit · bfloat — Brain floating point 16 — preferred for training	Does not fit: needs 32.5 GB

Selected Q4_K_M

4-bit medium quantization

11.9GBat 16K context

Weights: 8.9 GB
KV cache: 2.7 GB
Overhead: 0.3 GB

Fits RTX 4090 (reference) · 12.2 GB free · ~69 tok/s

≥50 tok/s needs ≥730 GB/s · RTX 5090 ~123, RTX 3090 Ti ~69

Which Hardware Can Run Hermes 4 14B?

Q4_K_M · 9.5 GB

Hermes 4 14B (Q4_K_M) requires 9.5 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 13+ GB is recommended. Using the full 41K context window can add up to 6.4 GB, bringing total usage to 15.9 GB. 40 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 3080 Ti. 150 devices with unified memory can run Hermes 4 14B, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Mini M2 Pro (16 GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M

Typical hardwareQuantVRAMtok/sContext that fits

8 GB · RTX 4060IQ2_M · 8 GB · ~32 tok/s · 16K8~3216K8 GB tier →

12 GB · RTX 3060 12GBQ4_K_M · 11.9 GB · ~25 tok/s · 16K11.9~2516K12 GB tier →

16 GB · RTX 5060 Ti 16GBQ4_K_M · 11.9 GB · ~31 tok/s · 40K11.9~3140K16 GB tier →

24 GB · RTX 4090referenceQ4_K_M · 11.9 GB · ~69 tok/s · 40K11.9~6940K24 GB tier →

Mac 32 GB · Mac Mini M4Q4_K_M · 11.9 GB · ~9 tok/s · 40K11.9~940KMacs →

Mac 64 GB · Mac Studio M4 MaxQ4_K_M · 11.9 GB · ~40 tok/s · 40K11.9~4040KMacs →

Alternatives to Hermes 4 14B

Hermes family · 4 models →

Similar models

6tok/s on RTX 4090

Model	Score	VRAM	tok/s	Why it is similar
Apriel 1.6 15B ThinkerServiceNow-AI	—	13.2 GB	~67	same size
Phi 4 ReasoningMicrosoft	—	12.5 GB	~69	same size
Phi 3 Medium 128k InstructMicrosoft	—	12 GB	~72	same size
Qwen1.5 14BAlibaba	—	22.2 GB	~63	same size
JiRackUltra 14BCMSManhattan	—	12.4 GB	~69	same size
Nemotron Terminal 14BNVIDIA	—	11.9 GB	~69	same size

Specifications

Publisher: Nous Research
Family: Hermes
Parameters: 14.8B
Architecture: Qwen3ForCausalLM
Context Length: 40,960 tokens
Vocabulary Size: 151,936
Release Date: 2025-08-30
License: Apache 2.0

Resources

Frequently Asked Questions

How much VRAM does Hermes 4 14B need?

Hermes 4 14B requires 9.5 GB of VRAM at Q4_K_M, or 30.2 GB at BF16. Full 41K context adds up to 6.4 GB (15.9 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 14.8B × 4.8 bits ÷ 8 = 8.9 GB

KV Cache + Overhead ≈ 0.6 GB (at 2K context + ~0.3 GB framework)

Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.

KV Cache + Overhead ≈ 7 GB (at full 41K context)

VRAM usage by quantization

Q4_K_M

9.5 GB

Q4_K_M + full context

15.9 GB

Learn more about VRAM estimation →

Can NVIDIA GeForce RTX 4090 run Hermes 4 14B?

Yes, at Q8_0 (15.4 GB) or lower. Higher quantizations like BF16 (30.2 GB) exceed the NVIDIA GeForce RTX 4090's 24 GB.

What's the best quantization for Hermes 4 14B?

For Hermes 4 14B, Q4_K_M (9.5 GB) offers the best balance of quality and VRAM usage. Q4_K_L (9.7 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XS at 5.1 GB.

VRAM requirement by quantization

IQ2_XS

5.1 GB~57%

Q3_K_S

7.1 GB~77%

IQ4_XS

8.6 GB~87%

Q4_K_M ★

9.5 GB~89%

Q5_K_S

10.8 GB~92%

BF16

30.2 GB~100%

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run Hermes 4 14B on a Mac?

Yes — Mac Mini M2 Pro (16 GB) and 133 other Macs can run Hermes 4 14B. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 9.5 GB of usable unified memory (RAM minus macOS overhead).

Can I run Hermes 4 14B locally?

Yes — Hermes 4 14B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is Hermes 4 14B?

At Q4_K_M, Hermes 4 14B can reach ~505 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~69 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

tok/s = (bandwidth GB/s ÷ model GB) × efficiency

Example: NVIDIA B200 → 8000 ÷ 9.5 × 0.65 = ~547 tok/s

Estimated speed at Q4_K_M (9.5 GB)

NVIDIA B200

~547 tok/s

NVIDIA GeForce RTX 4090

~69 tok/s

NVIDIA B300

~547 tok/s

AMD Instinct MI350X

~505 tok/s

Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.

Learn more about tok/s estimation →

What's the download size of Hermes 4 14B?

At Q4_K_M, the download is about 8.86 GB. The full-precision BF16 version is 29.54 GB. The smallest option (IQ2_XS) is 4.43 GB.

Which GPUs can run Hermes 4 14B?

40 consumer GPUs can run Hermes 4 14B at Q4_K_M. Top options include AMD Radeon RX 6800, AMD Radeon RX 6800 XT, AMD Radeon RX 6900 XT, AMD Radeon RX 6700 XT. 26 GPUs have plenty of headroom for comfortable inference.

Which devices can run Hermes 4 14B?

150 devices with unified memory can run Hermes 4 14B at Q4_K_M (9.5 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.