Qwen · Over 80B · Mixture of experts
Qwen3 235B A22B
235.09 billion parameters in total, but only 22.14 billion are read for each token. That split is the whole point of the design: it costs the memory of a large model and the speed of a small one. It uses grouped-query attention with 4 KV heads for 64 query heads, which cuts the cache to 6% of what full attention would need.
About Qwen3 235B A22B
The flagship Qwen3 mixture-of-experts model: 235B parameters, 22B active. It is a server-class model, and it is here because people ask whether it runs at home.
Qwen3 235B A22B is a mixture-of-experts model with 235.09B parameters, 22.14B of them active for each token, built from 94 layers with a hidden size of 4096. Its repository was published on 2025-04-27. The license listed on Hugging Face is apache-2.0. It uses grouped-query attention with 4 KV heads for 64 query heads, which cuts the cache to 6% of what full attention would need.
Memory needed, by quantization
At an 8k context window. Quality is our rough ranking of how much the compression costs you: anything at or above Q5 is hard to tell apart from the original in normal use.
| Quantization | File size | KV cache | Total VRAM | Quality | What it costs you |
|---|---|---|---|---|---|
| Q8_0 | 232.63 GB | 1.47 GB | 233.89 GB | 99% | Lossless in practice. Use it when the memory is there. |
| Q6_K | 179.53 GB | 1.47 GB | 180.93 GB | 98% | Very close to Q8 for two thirds of the size. |
| Q5_K_M | 155.72 GB | 1.47 GB | 157.19 GB | 96% | The quality-first choice when Q6 will not fit. |
| Q4_K_M | 132.19 GB | 1.47 GB | 133.71 GB | 93% | The default. Best size-to-quality ratio for local use. |
| IQ4_XS | 116.31 GB | 1.47 GB | 117.88 GB | 91% | Importance-matrix 4-bit. Q4_K_S quality, smaller file. |
| Q3_K_M | 107.01 GB | 1.47 GB | 108.6 GB | 86% | Degradation starts to show. A way to fit one size up. |
| IQ3_XXS | 83.75 GB | 1.47 GB | 85.4 GB | 79% | Aggressive. Only worth it on very large models. |
What a longer conversation costs
Same model at Q4_K_M, only the context window changes. The cache grows in a straight line with every token in the window.
| Context | KV cache | Total VRAM | Fits in 8 GB | Fits in 12 GB | Fits in 24 GB |
|---|---|---|---|---|---|
| 4k | 0.73 GB | 132.98 GB | no | no | no |
| 8k | 1.47 GB | 133.71 GB | no | no | no |
| 16k | 2.94 GB | 135.18 GB | no | no | no |
| 32k | 5.88 GB | 138.12 GB | no | no | no |
The "fits" columns allow for the roughly 0.8 GB Windows keeps for the desktop.
Which hardware runs Qwen3 235B A22B
1 of 118 consumer devices run it at a quantization worth using, at an 8k context window.
| Device | Memory | Verdict | Quantization | Used | Speed |
|---|---|---|---|---|---|
| Apple M3 Ultra 512GB Apple | 512 GB | Runs well | Q8_0 | 234.47 GB | 13.8 t/s |
| Apple M1 Ultra 64GB Apple | 64 GB | Just fits | n/a ! | 58.22 GB | 46.6 t/s |
| Apple M2 Ultra 64GB Apple | 64 GB | Just fits | n/a ! | 58.22 GB | 46.6 t/s |
| Apple M3 Ultra 96GB Apple | 96 GB | Just fits | n/a ! | 85.59 GB | 34.6 t/s |
| Apple M4 Max 64GB Apple | 64 GB | Just fits | n/a ! | 58.22 GB | 31.8 t/s |
| Apple M1 Ultra 128GB Apple | 128 GB | Just fits | n/a | 108.85 GB | 27.4 t/s |
| Apple M2 Ultra 128GB Apple | 128 GB | Just fits | n/a | 108.85 GB | 27.4 t/s |
| Apple M1 Max 64GB Apple | 64 GB | Just fits | n/a ! | 58.22 GB | 23.3 t/s |
| Apple M2 Max 64GB Apple | 64 GB | Just fits | n/a ! | 58.22 GB | 23.3 t/s |
| Apple M3 Max 64GB Apple | 64 GB | Just fits | n/a ! | 58.22 GB | 23.3 t/s |
| Apple M2 Ultra 192GB Apple | 192 GB | Just fits | n/a | 157.57 GB | 19.6 t/s |
| Apple M4 Max 128GB Apple | 128 GB | Just fits | n/a | 108.85 GB | 18.7 t/s |
| Apple M2 Max 96GB Apple | 96 GB | Just fits | n/a ! | 85.59 GB | 16.9 t/s |
| Apple M3 Max 96GB Apple | 96 GB | Just fits | n/a ! | 85.59 GB | 16.9 t/s |
| Apple M4 Pro 64GB Apple | 64 GB | Just fits | n/a ! | 58.22 GB | 15.9 t/s |
| Apple M3 Ultra 256GB Apple | 256 GB | Just fits | n/a | 234.47 GB | 13.8 t/s |
| Apple M3 Max 128GB Apple | 128 GB | Just fits | n/a | 108.85 GB | 13.7 t/s |
| Jetson AGX Orin 64GB NVIDIA | 64 GB | Just fits | n/a ! | 58.25 GB | 12.9 t/s |
| Ryzen AI Max+ 395 64GB AMD | 64 GB | Just fits | n/a ! | 58.25 GB | 12.5 t/s |
| NVIDIA DGX Spark 128GB NVIDIA | 128 GB | Just fits | n/a | 108.88 GB | 10.1 t/s |
| Ryzen AI Max+ 395 96GB AMD | 96 GB | Just fits | n/a ! | 85.62 GB | 9 t/s |
| Ryzen AI Max+ 395 128GB AMD | 128 GB | Just fits | n/a | 108.88 GB | 7.3 t/s |
| RTX PRO 6000 Blackwell NVIDIA | 96 GB | Partial offload 94/94 | n/a | 133.71 GB | 10.5 t/s |
| RTX 6000 Ada Generation NVIDIA | 48 GB | Partial offload 94/94 | n/a | 133.71 GB | 5 t/s |
| RTX A6000 NVIDIA | 48 GB | Partial offload 94/94 | n/a | 133.71 GB | 4.9 t/s |
Show 35 more devices
| Radeon PRO W7900 AMD | 48 GB | Partial offload 94/94 | n/a | 133.71 GB | 4.8 t/s |
| GeForce RTX 5090 NVIDIA | 32 GB | Partial offload 94/94 | n/a | 133.71 GB | 4.5 t/s |
| GeForce RTX 4090 NVIDIA | 24 GB | Partial offload 94/94 | n/a | 133.71 GB | 4.1 t/s |
| GeForce RTX 3090 Ti NVIDIA | 24 GB | Partial offload 94/94 | n/a | 133.71 GB | 4.1 t/s |
| GeForce RTX 3090 NVIDIA | 24 GB | Partial offload 94/94 | n/a | 133.71 GB | 4 t/s |
| GeForce RTX 5090 Laptop NVIDIA | 24 GB | Partial offload 94/94 | n/a | 133.71 GB | 4 t/s |
| RTX A5000 NVIDIA | 24 GB | Partial offload 94/94 | n/a | 133.71 GB | 4 t/s |
| Radeon RX 7900 XTX AMD | 24 GB | Partial offload 94/94 | n/a | 133.71 GB | 4 t/s |
| GeForce RTX 5080 NVIDIA | 16 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.8 t/s |
| GeForce RTX 5070 Ti NVIDIA | 16 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.8 t/s |
| GeForce RTX 5080 Laptop NVIDIA | 16 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.8 t/s |
| Radeon RX 7900 XT AMD | 20 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.8 t/s |
| GeForce RTX 4080 SUPER NVIDIA | 16 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.7 t/s |
| GeForce RTX 4080 NVIDIA | 16 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.7 t/s |
| GeForce RTX 4070 Ti SUPER NVIDIA | 16 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.7 t/s |
| GeForce RTX 3080 Ti NVIDIA | 12 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.7 t/s |
| GeForce RTX 3080 12GB NVIDIA | 12 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.7 t/s |
| GeForce RTX 4090 Laptop NVIDIA | 16 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.7 t/s |
| Radeon RX 9070 XT AMD | 16 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.7 t/s |
| Radeon RX 9070 AMD | 16 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.7 t/s |
| GeForce RTX 5070 NVIDIA | 12 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.6 t/s |
| GeForce RTX 5060 Ti 16GB NVIDIA | 16 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.6 t/s |
| GeForce RTX 3080 10GB NVIDIA | 10 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.6 t/s |
| GeForce RTX 2080 Ti NVIDIA | 11 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.6 t/s |
| RTX A4000 NVIDIA | 16 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.6 t/s |
| Radeon RX 7900 GRE AMD | 16 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.6 t/s |
| Radeon RX 7800 XT AMD | 16 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.6 t/s |
| Radeon RX 6900 XT AMD | 16 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.6 t/s |
| Radeon RX 6800 AMD | 16 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.6 t/s |
| Arc A770 16GB Intel | 16 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.6 t/s |
| GeForce RTX 4070 Ti NVIDIA | 12 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.5 t/s |
| GeForce RTX 4070 SUPER NVIDIA | 12 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.5 t/s |
| GeForce RTX 4070 NVIDIA | 12 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.5 t/s |
| GeForce RTX 3070 Ti NVIDIA | 8 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.5 t/s |
| GeForce GTX 1080 Ti NVIDIA | 11 GB | Partial offload 94/94 | n/a | 133.71 GB | 3.5 t/s |
How it compares
Other models in the same family or of a similar size, on a 12 GB card and a 24 GB card, each at the best quantization that fits.
| Model | Weights at 4 bits | GeForce RTX 3060 12GB | GeForce RTX 4090 |
|---|---|---|---|
| Qwen3 235B A22B | 132.2 GB | does not fit well | does not fit well |
| Qwen3.5 122B A10B | 70.3 GB | 12.6 t/s at Q4_K_M | 17.7 t/s at Q4_K_M |
| Qwen3 Next 80B A3B Instruct | 45.7 GB | 32.7 t/s at Q4_K_M | 54.6 t/s at Q4_K_M |
| Qwen3 Coder Next | 44.8 GB | 33.2 t/s at Q4_K_M | 56.4 t/s at Q4_K_M |
| MiniMax M2.7 | 128.6 GB | does not fit well | 10.4 t/s at Q3_K_M |
| MiniMax M2.5 | 128.6 GB | does not fit well | 10.4 t/s at Q3_K_M |
What to buy to run Qwen3 235B A22B
The cheapest hardware that runs it at a quantization worth using and a speed you would not resent, at an 8k context window.
Cheapest that works
Apple M3 Ultra 256GB
$5,599 whole machine
Q6_K · uses 181.38 GB of its 256 GB · about 17.6 tokens/s
Best value
Apple M2 Ultra 192GB
$6,599 whole machine
Q5_K_M · about 19.6 tokens/s · 800 GB/s
If it just has to run
GeForce RTX 2060 12GB
$160 used
Q3_K_M · about 4.2 tokens/s. Cheaper because it runs the model at a lower precision.
| Device | Price | Memory | Runs it at | Speed | |
|---|---|---|---|---|---|
| Apple M3 Ultra 256GB Apple · whole machine | $5,599 | 256 GB | Q6_K | 17.6 t/s | Check price |
| Apple M2 Ultra 192GB Apple · whole machine | $6,599 | 192 GB | Q5_K_M | 19.6 t/s | Check price |
| Apple M3 Ultra 512GB Apple · whole machine | $9,499 | 512 GB | Q6_K | 17.6 t/s | Check price |
Indicative prices reviewed 2026-09-01; used prices are marketplace typical. Options that cost more than a cheaper one with no more memory and no more speed are hidden. Price links are Amazon affiliate links. Change the standard, the context window or the budget in the buying tool.
These numbers also cover
Fine-tunes share their base model's architecture, so they need exactly the same memory. If you are looking for one of these, the figures above apply unchanged.
Qwen/Qwen3-235B-A22B-Instruct-2507
Where it sits in the catalog
Source: Qwen/Qwen3-235B-A22B on Hugging Face. Downloaded 343k times in the last month. Published 2025-04-27. Architecture figures are read from the repository's own configuration file, so they move when the model does.
Qwen3.5 122B A10B
125.09B (8.17B active) · 256k context
Qwen3 Next 80B A3B Instruct
81.32B (3.19B active) · 256k context
Qwen3 Coder Next
79.67B (3.19B active) · 256k context
Qwen2.5 VL 72B Instruct
73.41B · 125k context
Qwen2.5 72B Instruct
72.71B · 32k context
Qwen3.5 35B A3B
35.95B (2.9B active) · 256k context
Qwen3 32B
32.76B · 40k context
Qwen2.5 32B Instruct
32.76B · 32k context
Questions people ask about Qwen3 235B A22B
How much VRAM does Qwen3 235B A22B need?
At Q4_K_M with an 8k context window, Qwen3 235B A22B needs about 133.71 GB: 131.84 GB of weights, 1.47 GB of KV cache and 0.41 GB of overhead. At the highest-quality file it needs about 233.89 GB.
What is the cheapest GPU that runs Qwen3 235B A22B?
The smallest consumer device in our catalog that fits it is the Apple M3 Ultra 512GB with 512 GB, running it at Q8_0 at roughly 13.8 tokens per second.
How much does a longer context window add for Qwen3 235B A22B?
On the memory side, a 32k window adds 5.9 GB of KV cache on top of the weights, and the cache grows in a straight line with every token. The longest window the model supports natively is 40k tokens.