Qwen · 9B to 32B · Mixture of experts
Qwen3 30B A3B
30.53 billion parameters in total, but only 3.34 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 32 query heads, which cuts the cache to 13% of what full attention would need.
About Qwen3 30B A3B
A mixture-of-experts Qwen3 with 128 experts, 8 of them active for each token. It needs the memory of a 30B model and generates at roughly the speed of a 3B one, which makes it a favorite wherever the whole model fits in fast memory, and a good candidate for expert offload where it does not.
Qwen3 30B A3B is a mixture-of-experts model with 30.53B parameters, 3.34B of them active for each token, built from 48 layers with a hidden size of 2048. 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 32 query heads, which cuts the cache to 13% 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 | 30.21 GB | 0.75 GB | 30.99 GB | 99% | Lossless in practice. Use it when the memory is there. |
| Q6_K | 23.32 GB | 0.75 GB | 24.17 GB | 98% | Very close to Q8 for two thirds of the size. |
| Q5_K_M | 20.22 GB | 0.75 GB | 21.11 GB | 96% | The quality-first choice when Q6 will not fit. |
| Q4_K_M | 17.17 GB | 0.75 GB | 18.08 GB | 93% | The default. Best size-to-quality ratio for local use. |
| IQ4_XS | 15.11 GB | 0.75 GB | 16.04 GB | 91% | Importance-matrix 4-bit. Q4_K_S quality, smaller file. |
| Q3_K_M | 13.9 GB | 0.75 GB | 14.84 GB | 86% | Degradation starts to show. A way to fit one size up. |
| IQ3_XXS | 10.88 GB | 0.75 GB | 11.85 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 | 0.38 GB | 17.71 GB | no | no | yes |
| 8k | 0.75 GB | 18.08 GB | no | no | yes |
| 16k | 1.5 GB | 18.83 GB | no | no | yes |
| 32k | 3 GB | 20.33 GB | no | no | yes |
The "fits" columns allow for the roughly 0.8 GB Windows keeps for the desktop.
Which hardware runs Qwen3 30B A3B
55 of 118 consumer devices run it at a quantization worth using, at an 8k context window.
| Device | Memory | Verdict | Quantization | Used | Speed |
|---|---|---|---|---|---|
| GeForce RTX 5090 NVIDIA | 32 GB | Just fits | Q8_0 | 30.99 GB | 190.2 t/s |
| RTX PRO 6000 Blackwell NVIDIA | 96 GB | Runs great | Q8_0 | 30.99 GB | 190.2 t/s |
| RTX 6000 Ada Generation NVIDIA | 48 GB | Runs great | Q8_0 | 30.99 GB | 101.9 t/s |
| RTX A6000 NVIDIA | 48 GB | Runs great | Q8_0 | 30.99 GB | 81.5 t/s |
| Apple M3 Ultra 96GB Apple | 96 GB | Runs great | Q8_0 | 31.27 GB | 80.3 t/s |
| Apple M3 Ultra 256GB Apple | 256 GB | Runs great | Q8_0 | 31.27 GB | 80.3 t/s |
| Apple M3 Ultra 512GB Apple | 512 GB | Runs great | Q8_0 | 31.27 GB | 80.3 t/s |
| Apple M1 Ultra 64GB Apple | 64 GB | Runs great | Q8_0 | 31.27 GB | 78.4 t/s |
| Apple M1 Ultra 128GB Apple | 128 GB | Runs great | Q8_0 | 31.27 GB | 78.4 t/s |
| Apple M2 Ultra 64GB Apple | 64 GB | Runs great | Q8_0 | 31.27 GB | 78.4 t/s |
| Apple M2 Ultra 128GB Apple | 128 GB | Runs great | Q8_0 | 31.27 GB | 78.4 t/s |
| Apple M2 Ultra 192GB Apple | 192 GB | Runs great | Q8_0 | 31.27 GB | 78.4 t/s |
| Radeon PRO W7900 AMD | 48 GB | Runs great | Q8_0 | 30.99 GB | 77.8 t/s |
| Apple M4 Max 48GB Apple | 48 GB | Runs great | Q8_0 | 31.27 GB | 53.5 t/s |
| Apple M4 Max 64GB Apple | 64 GB | Runs great | Q8_0 | 31.27 GB | 53.5 t/s |
| Apple M4 Max 128GB Apple | 128 GB | Runs great | Q8_0 | 31.27 GB | 53.5 t/s |
| Apple M1 Max 64GB Apple | 64 GB | Runs great | Q8_0 | 31.27 GB | 39.2 t/s |
| Apple M2 Max 64GB Apple | 64 GB | Runs great | Q8_0 | 31.27 GB | 39.2 t/s |
| Apple M2 Max 96GB Apple | 96 GB | Runs great | Q8_0 | 31.27 GB | 39.2 t/s |
| Apple M3 Max 48GB Apple | 48 GB | Runs great | Q8_0 | 31.27 GB | 39.2 t/s |
| Apple M3 Max 64GB Apple | 64 GB | Runs great | Q8_0 | 31.27 GB | 39.2 t/s |
| Apple M3 Max 96GB Apple | 96 GB | Runs great | Q8_0 | 31.27 GB | 39.2 t/s |
| Apple M3 Max 128GB Apple | 128 GB | Runs great | Q8_0 | 31.27 GB | 39.2 t/s |
| NVIDIA DGX Spark 128GB NVIDIA | 128 GB | Runs great | Q8_0 | 31.3 GB | 29 t/s |
| Apple M4 Pro 48GB Apple | 48 GB | Runs great | Q8_0 | 31.27 GB | 26.8 t/s |
Show 35 more devices
| Apple M4 Pro 64GB Apple | 64 GB | Runs great | Q8_0 | 31.27 GB | 26.8 t/s |
| Jetson AGX Orin 64GB NVIDIA | 64 GB | Runs well | Q8_0 | 31.3 GB | 21.7 t/s |
| Ryzen AI Max+ 395 64GB AMD | 64 GB | Runs well | Q8_0 | 31.3 GB | 21 t/s |
| Ryzen AI Max+ 395 96GB AMD | 96 GB | Runs well | Q8_0 | 31.3 GB | 21 t/s |
| Ryzen AI Max+ 395 128GB AMD | 128 GB | Runs well | Q8_0 | 31.3 GB | 21 t/s |
| GeForce RTX 4090 NVIDIA | 24 GB | Runs great | Q5_K_M | 21.11 GB | 146.3 t/s |
| GeForce RTX 3090 Ti NVIDIA | 24 GB | Runs great | Q5_K_M | 21.11 GB | 146.3 t/s |
| GeForce RTX 3090 NVIDIA | 24 GB | Runs great | Q5_K_M | 21.11 GB | 135.9 t/s |
| GeForce RTX 5090 Laptop NVIDIA | 24 GB | Runs great | Q5_K_M | 21.11 GB | 130.1 t/s |
| Radeon RX 7900 XTX AMD | 24 GB | Runs great | Q5_K_M | 21.11 GB | 118.2 t/s |
| RTX A5000 NVIDIA | 24 GB | Runs great | Q5_K_M | 21.11 GB | 111.5 t/s |
| Radeon RX 7900 XT AMD | 20 GB | Runs great | Q4_K_M | 18.08 GB | 111 t/s |
| GeForce RTX 5080 NVIDIA | 16 GB | Just fits | Q3_K_M | 14.84 GB | 181.7 t/s |
| GeForce RTX 5070 Ti NVIDIA | 16 GB | Just fits | Q3_K_M | 14.84 GB | 169.6 t/s |
| GeForce RTX 5080 Laptop NVIDIA | 16 GB | Just fits | Q3_K_M | 14.84 GB | 145.3 t/s |
| GeForce RTX 4080 SUPER NVIDIA | 16 GB | Just fits | Q3_K_M | 14.84 GB | 139.3 t/s |
| GeForce RTX 4080 NVIDIA | 16 GB | Just fits | Q3_K_M | 14.84 GB | 135.7 t/s |
| GeForce RTX 4070 Ti SUPER NVIDIA | 16 GB | Just fits | Q3_K_M | 14.84 GB | 127.2 t/s |
| GeForce RTX 4090 Laptop NVIDIA | 16 GB | Just fits | Q3_K_M | 14.84 GB | 109 t/s |
| Radeon RX 9070 XT AMD | 16 GB | Just fits | Q3_K_M | 14.84 GB | 103.5 t/s |
| Radeon RX 9070 AMD | 16 GB | Just fits | Q3_K_M | 14.84 GB | 103.5 t/s |
| Radeon RX 7800 XT AMD | 16 GB | Just fits | Q3_K_M | 14.84 GB | 100.2 t/s |
| Radeon RX 7900 GRE AMD | 16 GB | Just fits | Q3_K_M | 14.84 GB | 92.5 t/s |
| GeForce RTX 5060 Ti 16GB NVIDIA | 16 GB | Just fits | Q3_K_M | 14.84 GB | 84.8 t/s |
| RTX A4000 NVIDIA | 16 GB | Just fits | Q3_K_M | 14.84 GB | 84.8 t/s |
| Radeon RX 6900 XT AMD | 16 GB | Just fits | Q3_K_M | 14.84 GB | 82.2 t/s |
| Radeon RX 6800 AMD | 16 GB | Just fits | Q3_K_M | 14.84 GB | 82.2 t/s |
| Arc A770 16GB Intel | 16 GB | Just fits | Q3_K_M | 14.84 GB | 81.8 t/s |
| GeForce RTX 4060 Ti 16GB NVIDIA | 16 GB | Just fits | Q3_K_M | 14.84 GB | 54.5 t/s |
| Radeon RX 7600 XT AMD | 16 GB | Just fits | Q3_K_M | 14.84 GB | 46.2 t/s |
| Apple M4 Max 36GB Apple | 36 GB | Just fits | n/a | 31.27 GB | 53.5 t/s |
| Apple M1 Max 32GB Apple | 32 GB | Just fits | n/a | 24.37 GB | 48.2 t/s |
| Apple M2 Max 32GB Apple | 32 GB | Just fits | n/a | 24.37 GB | 48.2 t/s |
| Apple M1 Pro 16GB Apple | 16 GB | Just fits | n/a ! | 11.93 GB | 40.9 t/s |
| Apple M2 Pro 16GB Apple | 16 GB | Just fits | n/a ! | 11.93 GB | 40.9 t/s |
Can my GPU run Qwen3 30B A3B?
A direct answer for popular devices, each with the settings we would use.
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 30B A3B | 17.2 GB | 36.3 t/s at Q4_K_M | 146.3 t/s at Q5_K_M |
| Qwen3 Coder 30B A3B Instruct | 17.2 GB | 36.3 t/s at Q4_K_M | 146.3 t/s at Q5_K_M |
| Qwen3 32B | 18.4 GB | does not fit well | 37.1 t/s at Q4_K_M |
| Qwen2.5 32B Instruct | 18.4 GB | does not fit well | 37.1 t/s at Q4_K_M |
| Gemma 4 31B Instruct | 17.6 GB | does not fit well | 35.5 t/s at Q4_K_M |
| Gemma 4 26B A4B Instruct | 14.5 GB | 37.6 t/s at Q4_K_M | 117 t/s at Q6_K |
What to buy to run Qwen3 30B A3B
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
GeForce GTX 1650
$100 used
Q4_K_M · uses 18.08 GB of its 4 GB · about 17 tokens/s
Best value
Radeon RX 7900 XT
$650
Q4_K_M · about 111 tokens/s · 800 GB/s
| Device | Price | Memory | Runs it at | Speed | |
|---|---|---|---|---|---|
| GeForce GTX 1650 NVIDIA | $100 used | 4 GB | Q4_K_M | 17 t/s | Check price |
| GeForce GTX 1660 SUPER NVIDIA | $110 used | 6 GB | Q4_K_M | 26 t/s | Check price |
| GeForce GTX 1080 Ti NVIDIA | $150 used | 11 GB | Q4_K_M | 38.5 t/s | Check price |
| GeForce RTX 2060 12GB NVIDIA | $160 used | 12 GB | Q4_K_M | 35.1 t/s | Check price |
| GeForce RTX 2080 Ti NVIDIA | $250 used | 11 GB | Q4_K_M | 41.8 t/s | Check price |
| Arc A770 16GB Intel | $280 used | 16 GB | Q4_K_M | 54.5 t/s | Check price |
| Radeon RX 6800 AMD | $330 used | 16 GB | Q4_K_M | 54.7 t/s | Check price |
| GeForce RTX 5060 Ti 16GB NVIDIA | $430 | 16 GB | Q4_K_M | 55.8 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/Qwen3-30B-A3B-Instruct-2507mlabonne/Qwen3-30B-A3B-abliteratedQwen/Qwen3-30B-A3B-Basebineric/lynx-instruct-30b
Where it sits in the catalog
Source: Qwen/Qwen3-30B-A3B on Hugging Face. Downloaded 2.4M 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 Coder 30B A3B Instruct
30.53B (3.34B active) · 40k context
Qwen3 32B
32.76B · 40k context
Qwen2.5 32B Instruct
32.76B · 32k context
Qwen2.5 Coder 32B Instruct
32.76B · 32k context
Qwen3.5 27B
27.78B · 256k context
Qwen3.5 35B A3B
35.95B (2.9B active) · 256k context
Qwen3 14B
14.77B · 40k context
Qwen2.5 14B Instruct
14.77B · 32k context
Questions people ask about Qwen3 30B A3B
How much VRAM does Qwen3 30B A3B need?
At Q4_K_M with an 8k context window, Qwen3 30B A3B needs about 18.08 GB: 16.99 GB of weights, 0.75 GB of KV cache and 0.34 GB of overhead. At the highest-quality file it needs about 30.99 GB.
What is the cheapest GPU that runs Qwen3 30B A3B?
The smallest consumer device in our catalog that fits it is the GeForce RTX 5080 with 16 GB, running it at Q3_K_M at roughly 181.7 tokens per second.
How much does a longer context window add for Qwen3 30B A3B?
On the memory side, a 32k window adds 3.0 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.