empero-aiQwen 3.8Qwen3_5ForConditionalGeneration

Qwen3.8 4B Distill — Hardware Requirements & GPU Compatibility

ChatReasoningFunctions

Qwen3.8 4B Distill is a 4.7B-parameter open language model from empero-ai in the Qwen 3.8 family. It supports a context window of up to 262,144 tokens. At Q4_K_M it needs about 3.26 GB of VRAM — see which GPUs and Macs can run it below.

4.7B params · dense262K contextApache 2.0Aug 2026137.0K downloads71 likes3.7K quant downloads

Based on Qwen3.5 4B

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 (Q4_K_M · 128K)Best on 24 GB (Q4_K_M · ~201 tok/s · 256K)Macs from 8 GB (Q4_K_M · ~22 tok/s)

Apache 2.0

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run Qwen3.8 4B Distill

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 RTX 3060 12GB (8 GB headroom) · Q4_K_M

Generation speed: ~72 tok/s; generation speed
Cost per 1M output tokens: $0.27; per 1M output tokens

Deployon Vast.aiBest optionfrom $0.070/hr

Compare GPUs →

Run Qwen3.8 4B Distill

3.4 GB VRAM at 16K context · fits the RTX 4090 (20.6 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/mradermacher/Qwen3.8-4B-Distill-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/mradermacher/Qwen3.8-4B-Distill-GGUF, then pick Q4_K_M. Or use the CLI (run LM Studio once first):
$lms get https://huggingface.co/mradermacher/Qwen3.8-4B-Distill-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 mradermacher/Qwen3.8-4B-Distill-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/mradermacher/Qwen3.8-4B-Distill-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 mradermacher/Qwen3.8-4B-Distill-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 empero-ai/Qwen3.8-4B-Distill --max-model-len 16384
Runs the original weights, not the Q4_K_M GGUF; needs ~10 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 SiddhJagani/Qwen3.8-4B-mlx-4Bit
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 SiddhJagani/Qwen3.8-4B-mlx-4Bit

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.
$mradermacher/Qwen3.8-4B-Distill-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
RTX 3060 12GBCheapestCheapest8 GB free	~72 tok/s	$0.27	Deploy on Vast.aifrom $0.070/hr (opens in new tab)	8 GB free · ~72 tok/s · $0.27 / 1M tokens
RTX A4000 (16 GB)12 GB free	~89 tok/s	$0.34	Deploy on Vast.aifrom $0.11/hr (opens in new tab)	16 GB · 12 GB free · ~89 tok/s · $0.34 / 1M tokens
RTX 3070 (8 GB)4 GB free	~89 tok/s	$0.40	Deploy on RunPodfrom $0.13/hr (opens in new tab)	8 GB · 4 GB free · ~89 tok/s · $0.40 / 1M tokens
RTX 3060 Ti (8 GB)4 GB free	~89 tok/s	$0.47	Deploy on Vast.aifrom $0.15/hr (opens in new tab)	8 GB · 4 GB free · ~89 tok/s · $0.47 / 1M tokens
RTX A5000 (24 GB)20 GB free	~153 tok/s	$0.29	Deploy on RunPodfrom $0.16/hr (opens in new tab)	24 GB · 20 GB free · ~153 tok/s · $0.29 / 1M tokens
RTX 5060 Ti 16GB12 GB free	~89 tok/s	$0.50	Deploy on Vast.aifrom $0.16/hr (opens in new tab)	12 GB free · ~89 tok/s · $0.50 / 1M tokens
RTX 3080 Ti (12 GB)8 GB free	~182 tok/s	$0.27	Deploy on RunPodfrom $0.18/hr (opens in new tab)	12 GB · 8 GB free · ~182 tok/s · $0.27 / 1M tokens
RTX 3090 (24 GB)20 GB free	~187 tok/s	$0.33	Deploy on RunPodfrom $0.22/hr (opens in new tab)	24 GB · 20 GB free · ~187 tok/s · $0.33 / 1M tokens

Compare GPUs →

How Much VRAM Does Qwen3.8 4B Distill Need?

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

VRAM	+32K	+128K	File	Quality	On RTX 4090 (reference)
2.5 GB	3 GB	5 GB	2 GB	2-bit · K-quant — 2-bit quantization with K-quant improvements	Fits: 256K context · ~267 tok/s
2.5 GB	3 GB	5 GB	2.2 GB	3-bit · i-quant small — Importance-weighted 3-bit, small	Fits: 256K context · ~267 tok/s
2.5 GB	3 GB	5 GB	2.1 GB	3-bit small — 3-bit small quantization	Fits: 256K context · ~261 tok/s
2.6 GB	3.1 GB	5.1 GB	2.2 GB	3-bit · i-quant medium — Importance-weighted 3-bit, medium	Fits: 256K context · ~256 tok/s
2.7 GB	3.2 GB	5.3 GB	2.3 GB	3-bit medium — 3-bit medium quantization	Fits: 256K context · ~239 tok/s
2.8 GB	3.3 GB	5.3 GB	2.6 GB	4-bit · legacy — 4-bit legacy quantization	Fits: 256K context · ~234 tok/s
2.9 GB	3.4 GB	5.4 GB	2.5 GB	3-bit large — 3-bit large quantization	Fits: 256K context · ~229 tok/s
3.0 GB	3.5 GB	5.5 GB	2.6 GB	4-bit · i-quant compact — Importance-weighted 4-bit, compact	Fits: 256K context · ~221 tok/s
3.1 GB	3.6 GB	5.6 GB	2.7 GB	4-bit · i-quant non-linear — Importance-weighted 4-bit, non-linear	Fits: 256K context · ~212 tok/s
3.1 GB	3.6 GB	5.6 GB	2.6 GB	4-bit small — 4-bit small quantization	Fits: 256K context · ~212 tok/s
3.1 GB	3.6 GB	5.6 GB	2.8 GB	4-bit · legacy+ — 4-bit legacy quantization with offset	Fits: 256K context · ~212 tok/s
3.3 GB	3.8 GB	5.8 GB	2.8 GB	4-bit medium — 4-bit medium quantization — most popular sweet spot	Fits: 256K context · ~201 tok/s
3.7 GB	4.2 GB	6.2 GB	3.1 GB	5-bit small — 5-bit small quantization	Fits: 256K context · ~179 tok/s
3.8 GB	4.3 GB	6.3 GB	3.2 GB	5-bit medium — 5-bit medium quantization — good quality/size tradeoff	Fits: 256K context · ~173 tok/s
4.3 GB	4.8 GB	6.8 GB	3.6 GB	6-bit, very good — 6-bit quantization, very good quality	Fits: 256K context · ~152 tok/s
5.1 GB	5.6 GB	7.7 GB	4.6 GB	8-bit, near-lossless — 8-bit quantization, near-lossless	Fits: 256K context · ~128 tok/s
9.8 GB	10.3 GB	12.3 GB	8.7 GB	16-bit · bfloat — Brain floating point 16 — preferred for training	Fits: 256K context · ~67 tok/s

Selected Q4_K_M

4-bit medium quantization

3.4GBat 16K context

Weights: 2.8 GB
KV cache: 0.3 GB
Overhead: 0.3 GB

Fits RTX 4090 (reference) · 20.6 GB free · ~201 tok/s

≥50 tok/s needs ≥251 GB/s · RTX 5090 ~357, RTX 3090 Ti ~201

Which Hardware Can Run Qwen3.8 4B Distill?

Q4_K_M · 3.3 GB

Qwen3.8 4B Distill (Q4_K_M) requires 3.3 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 5+ GB is recommended. Using the full 262K context window can add up to 5.2 GB, bringing total usage to 8.5 GB. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 163 devices with unified memory can run Qwen3.8 4B Distill, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), MacBook Air 13" M3 (8 GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M

Typical hardwareQuantVRAMtok/sContext that fits

8 GB · RTX 4060Q4_K_M · 3.4 GB · ~54 tok/s · 128K3.4~54128K8 GB tier →

12 GB · RTX 3060 12GBQ4_K_M · 3.4 GB · ~72 tok/s · 256K3.4~72256K12 GB tier →

16 GB · RTX 5060 Ti 16GBQ4_K_M · 3.4 GB · ~89 tok/s · 256K3.4~89256K16 GB tier →

24 GB · RTX 4090referenceQ4_K_M · 3.4 GB · ~201 tok/s · 256K3.4~201256K24 GB tier →

Mac 32 GB · Mac Mini M4Q4_K_M · 3.4 GB · ~26 tok/s · 256K3.4~26256KMacs →

Mac 64 GB · Mac Studio M4 MaxQ4_K_M · 3.4 GB · ~117 tok/s · 256K3.4~117256KMacs →

Alternatives to Qwen3.8 4B Distill

Qwen 3.8 family · 12 models →

Similar models

6tok/s on RTX 4090

Model	Score	VRAM	tok/s	Why it is similar
FrogNano 4B 2609Microsoft	—	3.4 GB	~201	same size
Qwen3.5 4BAlibaba	—	3.4 GB	~201	same size
Translategemma 4B ITGoogle	—	5.5 GB	~200	same size
Molmo2 4BAllen AI	—	4.7 GB	~193	same size
Qwen3 VL 4B InstructAlibaba	—	4.5 GB	~208	same size
Agents A1 4BInternScience	—	3.4 GB	~205	same size

Specifications

Publisher: empero-ai
Family: Qwen 3.8
Parameters: 4.7B
Architecture: Qwen3_5ForConditionalGeneration
Context Length: 262,144 tokens
Vocabulary Size: 248,320
Release Date: 2026-08-15
License: Apache 2.0

Resources

Frequently Asked Questions

How much VRAM does Qwen3.8 4B Distill need?

Qwen3.8 4B Distill requires 3.3 GB of VRAM at Q4_K_M, or 9.8 GB at BF16. Full 262K context adds up to 5.2 GB (8.5 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 4.7B × 4.8 bits ÷ 8 = 2.8 GB

KV Cache + Overhead ≈ 0.5 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 ≈ 5.7 GB (at full 262K context)

VRAM usage by quantization

Q4_K_M

3.3 GB

Q4_K_M + full context

8.5 GB

Learn more about VRAM estimation →

What's the best quantization for Qwen3.8 4B Distill?

For Qwen3.8 4B Distill, Q4_K_M (3.3 GB) offers the best balance of quality and VRAM usage. Q5_K_S (3.7 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 2.5 GB.

VRAM requirement by quantization

Q2_K

2.5 GB~75%

Q3_K_M

2.7 GB~83%

IQ4_NL

3.1 GB~88%

Q4_K_M ★

3.3 GB~89%

Q5_K_S

3.7 GB~92%

BF16

9.8 GB~100%

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run Qwen3.8 4B Distill on a Mac?

Yes — Mac Mini M2 (8 GB) and 145 other Macs can run Qwen3.8 4B Distill. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 3.3 GB of usable unified memory (RAM minus macOS overhead).

Can I run Qwen3.8 4B Distill locally?

Yes — Qwen3.8 4B Distill can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is Qwen3.8 4B Distill?

At Q4_K_M, Qwen3.8 4B Distill can reach ~1472 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~201 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 ÷ 3.26 × 0.65 = ~1595 tok/s

Estimated speed at Q4_K_M (3.3 GB)

NVIDIA B200

~1595 tok/s

NVIDIA GeForce RTX 4090

~201 tok/s

NVIDIA B300

~1595 tok/s

AMD Instinct MI350X

~1472 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 Qwen3.8 4B Distill?

At Q4_K_M, the download is about 2.80 GB. The full-precision BF16 version is 9.32 GB. The smallest option (Q2_K) is 1.98 GB.

Which GPUs can run Qwen3.8 4B Distill?

52 consumer GPUs can run Qwen3.8 4B Distill at Q4_K_M. Top options include AMD Radeon RX 6700 XT, AMD Radeon RX 6800, AMD Radeon RX 6800 XT. 52 GPUs have plenty of headroom for comfortable inference.

Which devices can run Qwen3.8 4B Distill?

163 devices with unified memory can run Qwen3.8 4B Distill at Q4_K_M (3.3 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.