AlibabaQwen 3Qwen3MoeForCausalLM

Qwen3 235B A22B — Hardware Requirements & GPU Compatibility

ChatFunctions

Qwen3 235B A22B is the largest model in Alibaba Cloud's Qwen 3 series, a Mixture of Experts (MoE) model with 235 billion total parameters and approximately 22 billion active parameters per forward pass. The MoE architecture enables it to deliver performance competitive with the best available open-weight models while requiring significantly less compute per token than a comparably sized dense model. It supports hybrid thinking mode for flexible chain-of-thought reasoning. Due to its massive total parameter count, running Qwen3 235B A22B locally requires substantial VRAM to load all expert weights, typically needing multiple high-end professional GPUs even at reduced precision. In heavily quantized formats it becomes accessible on workstation-class multi-GPU setups. Released under the Apache 2.0 license.

235.1B params · MoE22B active41K contextApache 2.0Apr 2025295.6K downloads1.1K likes67.7K quant downloads

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.

Qwen3 235B A22B runs on a 192 GB+ Mac or a datacenter GPU

Best local pick → Mac Pro M2 Ultra (192 GB) · Q4_K_M · 142.9 GB · ~23 tok/s

Macs from 128 GB (Q3_K_L · ~26 tok/s)

Score 100#58 of 106 open#36 of 82 that fit ≤192 GBApache 2.0

Not set up for it?

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

llmrun Composite Score →

100.0

90% range 96.5–103.4
#58 of 106 open models
#36 of 82 that fit ≤192 GB

Benchmarks · tier chart ↓

Among 106 open modelsHow the Composite Score is built

Coding101.390% range 96–107
AgentsNot enough data yet
MathNot enough data yet
ScienceNot enough data yet
Reasoning100.890% range 94–108

Run Qwen3 235B A22B

One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.

Ways to runOther ways

each opens step-by-step

Run in cloud

Your own private copy · Q4_K_M · 96 GB GPU

The easiest way to try this model
Running in a few minutes, easy setup
Chat right here or connect any app

$5+ bonus for new users

Run Qwen3 235B A22B

142.9 GB VRAM at 16K context · runs on one GH200 Grace Hopper Superchip 144 GB

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/unsloth/Qwen3-235B-A22B-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 search qwen/qwen3-235b-a22b, then pick Q4_K_M. Or use the CLI (run LM Studio once first):
$lms get qwen/qwen3-235b-a22b@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 unsloth/Qwen3-235B-A22B-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/unsloth/Qwen3-235B-A22B-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 unsloth/Qwen3-235B-A22B-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. Splits the model over 8 GPUs of one server (tensor parallel).
$vllm serve Qwen/Qwen3-235B-A22B --max-model-len 16384 --tensor-parallel-size 8
Runs the original weights, not the Q4_K_M GGUF; needs ~517 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/Qwen3-235B-A22B-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 mlx-community/Qwen3-235B-A22B-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.
$unsloth/Qwen3-235B-A22B-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
B200 (192 GB)CheapestCheapest49 GB free	~129 tok/s	$12.86	Deploy on RunPodfrom $5.98/hr (opens in new tab)	192 GB · 49 GB free · ~129 tok/s · $12.86 / 1M tokens
B300 (288 GB)145 GB free	~129 tok/s	$14.92	Deploy on RunPodfrom $6.94/hr (opens in new tab)	288 GB · 145 GB free · ~129 tok/s · $14.92 / 1M tokens
B200 (192 GB)49 GB free	~129 tok/s	$16.13	Deploy on Vast.aifrom $7.50/hr (opens in new tab)	192 GB · 49 GB free · ~129 tok/s · $16.13 / 1M tokens

Compare GPUs →

Your own private copy of Qwen3 235B A22B · Q4_K_M · OpenAI-compatible API

The pod’s first start downloads ~141 GB, billed while it loads.

Open the pod on RunPod
Deploy on RunPodH200 · $5.29/hr (opens in new tab)
New to RunPod? Sign up and add credit first. If that GPU is gone, pick another with at least 96 GB.
2
Set the model and your key
Press Set overrides → Environment variables. The first variable is already there; set its value:
Key
LLAMA_ARG_HF_REPO
Value
unsloth/Qwen3-235B-A22B-GGUF:Q4_K_M
Press Add Environment variable for your key:
Key
LLAMA_API_KEY
Value
Make up your own key
A long password works. You’ll paste it below as the API key. Without it, anyone who finds the pod’s address can use it.Also set Container disk to 170 GB.Then Set overrides.
Deploy
Press Deploy Pod. Usually ready in 15–60 min.

Use it

RunPod → your pod → Connect → HTTP 8080, copy the link. The key is the one you made up in step 2.

Endpoint URL

API key

Chats and keys stay only in this browser.

https://<POD_ID>-8080.proxy.runpod.net/v1

RunPod → your pod → Connect → HTTP 8080 gives this address. The API key is the one from step 2.

from openai import OpenAI
client = OpenAI(base_url="https://<POD_ID>-8080.proxy.runpod.net/v1", api_key="YOUR_KEY")
r = client.chat.completions.create(
    model="any",
    messages=[{"role": "user", "content": "Hello!"}])
print(r.choices[0].message.content)

Works with any OpenAI client.

https://<POD_ID>-8080.proxy.runpod.net/v1

RunPod → your pod → Connect → HTTP 8080 gives this address. The API key is the one from step 2.

Free on Windows, macOS, Linux, iOS, Android and the web.

Open the web app ↗ (opens in a new tab)Download ↗ (opens in a new tab)

1. Settings → Model Provider → Add → Add Custom Provider
2. API Mode OpenAI API Compatible; paste the API key and the base URL as API Host
3. Model → Fetch, pick the model
4. New chat → pick the provider → start chatting

Free and open source for Windows, macOS and Linux.

Download Jan ↗ (opens in a new tab)

1.Settings → next to Model Providers press +
2.API format OpenAI-compatible; Base URL is the base URL above (with /v1); API key is your key; Create
3.If no model appears, click + in Models and type any
4.New chat → pick the model → start chatting

How Much VRAM Does Qwen3 235B A22B Need?

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

VRAM	+32K	+128K	File	Quality	Runs on
100.4 GB	103.4 GB	—	85.7 GB	2-bit · K-quant — 2-bit quantization with K-quant improvements	Runs on Instinct MI250X 128 GB
103.3 GB	106.3 GB	—	101.4 GB	3-bit small — 3-bit small quantization	Runs on Instinct MI250X 128 GB
115.1 GB	118.1 GB	—	112.4 GB	3-bit medium — 3-bit medium quantization	Runs on Instinct MI250X 128 GB
121.0 GB	123.9 GB	—	111.2 GB	3-bit large — 3-bit large quantization	Runs on Instinct MI250X 128 GB
126.9 GB	129.8 GB	—	125.5 GB	4-bit · i-quant compact — Importance-weighted 4-bit, compact	Runs on H200 NVL 141 GB
132.7 GB	135.7 GB	—	133.7 GB	4-bit small — 4-bit small quantization	Runs on H200 NVL 141 GB
132.7 GB	135.7 GB	—	147.2 GB	4-bit · legacy+ — 4-bit legacy quantization with offset	Runs on H200 NVL 141 GB
141.6 GB	144.5 GB	—	142.2 GB	4-bit medium — 4-bit medium quantization — most popular sweet spot	Runs on GH200 Grace Hopper Superchip 144 GB
147.4 GB	150.4 GB	—	161.9 GB	5-bit · legacy — 5-bit legacy quantization	Runs on B200 192 GB
162.1 GB	165.1 GB	—	161.9 GB	5-bit small — 5-bit small quantization	Runs on B200 192 GB
168 GB	171 GB	—	166.8 GB	5-bit medium — 5-bit medium quantization — good quality/size tradeoff	Runs on B200 192 GB
194.4 GB	197.4 GB	—	193 GB	6-bit, very good — 6-bit quantization, very good quality	Runs on Instinct MI325X 256 GB
235.6 GB	238.6 GB	—	249.9 GB	8-bit, near-lossless — 8-bit quantization, near-lossless	Runs on Instinct MI325X 256 GB
470.7 GB	473.6 GB	—	470.3 GB	16-bit · bfloat — Brain floating point 16 — preferred for training	Runs on 7× H100 SXM 80 GB

Selected Q4_K_M

4-bit medium quantization

142.9GBat 16K context

Weights: 141.1 GB
KV cache: 1.6 GB
Overhead: 0.3 GB

Server-class: ~142.9 GB at 16K context · one GH200 Grace Hopper Superchip 144 GB

Or .

Which Hardware Can Run Qwen3 235B A22B?

Q4_K_M · 141.6 GB

Qwen3 235B A22B (Q4_K_M) requires 141.6 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 185+ GB is recommended. Using the full 41K context window can add up to 3.8 GB, bringing total usage to 145.3 GB. No consumer GPU has enough memory. 4 devices with unified memory can run Qwen3 235B A22B, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Pro M2 Ultra (192 GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M

Consumer GPUs (up to 32 GB) and Macs up to 96 GB: too small for this model (needs ≥ 101.8 GB)8 GB tier →12 GB tier →16 GB tier →24 GB tier →Macs →

How good is Qwen3 235B A22B?

The llmrun Score merges 9 public benchmark boards into a single figure, where 100 represents the average for open models.

Score vs VRAM in the ≤192 GB tier

Qwen3 235B A22B
Other models
Best Score for its VRAM
x: VRAM (GB, log scale) · y: llmrun Score

#36 of 82 that fit ≤192 GB

Each dot is a model at the best quant that fits ≤192 GB. The line steps up to the best Score available at each amount of VRAM; a dot below it is beaten by a model that needs no more memory.

Full ≤192 GB ranking →

Benchmarks

GPQA Diamond
Epoch AI · Apr 2025Independent
70.7%
#34/100 open
Independent
LMArena Text
LMArena · Oct 2026Arena
1402.1
#48/188 open
Arena
MMLU-Pro
TIGER-LabIndependent
68.2%
#41/146 open
Independent
Aider Polyglot
Epoch AI · Apr 2025Independent
59.6%
#3/18 open
Independent
SimpleBench
Epoch AI · Apr 2025Independent
31.0%
#15/25 open
Independent
MATH Level 5
Epoch AI · Apr 2025Independent
68.9%
#8/39 open
Independent

All 11 Qwen3 235B A22B benchmarks

In the ≤192 GB class

82 models fit ≤192 GB

Same Score, only models that fit ≤192 GB at their best quant: the ranking a ≤192 GB owner actually chooses from.

#1
MiMo V2.6 Flash RLXiaomi
131.9
~229 tok/s · 186.5 GB
#2
Step 5 PreviewStepFun
128.0
~151 tok/s · 191.4 GB
#3
DeepSeek V4 Flash 0731DeepSeek
127.6
~181 tok/s · 183.0 GB
#4
GLM 5.3 FlashZ.ai
124.1
~228 tok/s · 185.1 GB
#5
Qwen3.8 27BAlibaba
122.0
~299 tok/s · 17.9 GB
#6
Inkling SmallThinking Machines
119.6
~399 tok/s · 162.7 GB

Full ≤192 GB ranking →B200 page →

Alternatives to Qwen3 235B A22B

Qwen 3 family · 79 models →

Official models with an llmrun Score, each answering one question; sizes are each model's default quant (usually Q4_K_M) with a 16K-token context.

Best alternative at this size
Inkling Small
Thinking Machines
- 19.6 higher Score
- 19.8 GB more VRAM
Score 119.6 · 162.7 GB
Compare with Qwen3 235B A22B Open Inkling Small
Other vendor
MiMo V2.6 Flash RL
Xiaomi
- 31.9 higher Score
- 43.6 GB more VRAM
Score 131.9 · 186.5 GB
Compare with Qwen3 235B A22B Open MiMo V2.6 Flash RL

Similar models

Model	Score	VRAM	tok/s	Why it is similar
DeepSeek V4 Flash 0731DeepSeek	127.6	183 GB	—needs > 24 GB	similar score
GLM 5.3 FlashZ.ai	124.1	205.2 GB	—needs > 24 GB	similar score
DeepSeek V4 FlashDeepSeek	115.3	175.1 GB	—needs > 24 GB	similar score
Qwen3.5 397B A17BAlibaba	112.8	242.6 GB	—needs > 24 GB	similar score
MiniMax M3MiniMax	112.6	258 GB	—needs > 24 GB	similar score
MiniMax M2.5MiniMax	111.8	139.6 GB	—needs > 24 GB	similar score
Qwen3.5 122B A10BAlibaba	109.4	75.5 GB	—needs > 24 GB	similar score
GLM 4.7Z.ai	109.1	217.9 GB	—needs > 24 GB	similar score
MiMo V2.5Xiaomi	108.3	187.6 GB	—needs > 24 GB	similar score
GLM 4.6Z.ai	103.1	216.9 GB	—needs > 24 GB	similar score

Fine-tunes & variants(1)community builds of Qwen3 235B A22B

Baichuan M3 235B

Specifications

Publisher: Alibaba
Family: Qwen 3
Parameters: 235.1B
Active parameters: 22.2B per token
Architecture: Qwen3MoeForCausalLM
Context Length: 40,960 tokens
Vocabulary Size: 151,936
Release Date: 2025-04-27
License: Apache 2.0

Resources

Frequently Asked Questions

How much VRAM does Qwen3 235B A22B need?

Qwen3 235B A22B requires 141.6 GB of VRAM at Q4_K_M, or 470.7 GB at BF16. Full 41K context adds up to 3.8 GB (145.3 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 235.1B × 4.8 bits ÷ 8 = 141.1 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 ≈ 4.2 GB (at full 41K context)

VRAM usage by quantization

Q4_K_M

141.6 GB

Q4_K_M + full context

145.3 GB

Learn more about VRAM estimation →

Can NVIDIA GeForce RTX 5090 run Qwen3 235B A22B?

No — Qwen3 235B A22B requires at least 100.4 GB at Q2_K, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.

What's the best quantization for Qwen3 235B A22B?

For Qwen3 235B A22B, Q4_K_M (141.6 GB) offers the best balance of quality and VRAM usage. Q5_0 (147.4 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 100.4 GB.

VRAM requirement by quantization

Q2_K

100.4 GB~75%

Q3_K_L

121.0 GB~86%

Q4_K_M ★

141.6 GB~89%

Q5_0

147.4 GB~90%

Q5_K_M

168.0 GB~92%

BF16

470.7 GB~100%

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run Qwen3 235B A22B on a Mac?

Yes — Mac Pro M2 Ultra (128 GB) and 14 other Macs can run Qwen3 235B A22B, the smallest at Q3_K_L. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 141.6 GB of usable unified memory (RAM minus macOS overhead), from Mac Pro M2 Ultra (192 GB).

Can I run Qwen3 235B A22B locally?

Yes — Qwen3 235B A22B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is Qwen3 235B A22B?

At Q4_K_M, Qwen3 235B A22B can reach ~46 tok/s on AMD Instinct MI350X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

tok/s = 1000 ÷ (active GB ÷ (bandwidth GB/s × efficiency) × 1000 + layers × routing ms)

Mixture-of-Experts: just the active experts get read for each token, along with a fixed routing cost per layer.

Example: NVIDIA B200 → 13.4 GB active ÷ (8000 × 0.65) = 2.57 ms, plus 94 layers × 0.055 ms = 5.17 ms, so 1000 ÷ 7.74 ms = ~129 tok/s

Estimated speed at Q4_K_M (141.6 GB)

NVIDIA B200

~129 tok/s

NVIDIA B300

~129 tok/s

NVIDIA GH200 Grace Hopper Superchip

~107 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 235B A22B?

At Q4_K_M, the download is about 141.06 GB. The full-precision BF16 version is 470.19 GB. The smallest option (Q2_K) is 99.91 GB.

Which GPUs can run Qwen3 235B A22B?

No single consumer GPU has enough VRAM to run Qwen3 235B A22B at Q4_K_M (141.6 GB). Multi-GPU or professional hardware is required.

Which devices can run Qwen3 235B A22B?

4 devices with unified memory can run Qwen3 235B A22B at Q4_K_M (141.6 GB), including Mac Pro M2 Ultra (192 GB), Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Studio M2 Ultra (192 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.

Qwen3 235B A22B — Hardware Requirements & GPU Compatibility

Qwen3 235B A22B runs on a 192 GB+ Mac or a datacenter GPU

llmrun Composite Score →

Run Qwen3 235B A22B

Recommended

Ways to runOther ways

Run in cloud

How Much VRAM Does Qwen3 235B A22B Need?

Selected Q4_K_M

Which Hardware Can Run Qwen3 235B A22B?

How good is Qwen3 235B A22B?

Score vs VRAM in the ≤192 GB tier

Benchmarks

In the ≤192 GB class

Alternatives to Qwen3 235B A22B

Inkling Small

MiMo V2.6 Flash RL

Similar models

Specifications

Resources

Frequently Asked Questions