Qwen · 32B to 80B · Mixture of experts
Qwen3.5 35B A3B
35.95 billion parameters in total, but only 2.9 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. Most of its layers are linear-attention layers that keep a small fixed state, so only 10 of 40 grow with the window and long contexts cost far less than on a conventional model.
About Qwen3.5 35B A3B
Qwen3.5 35B A3B is a mixture-of-experts model with 35.95B parameters, 2.9B of them active for each token, built from 40 layers with a hidden size of 2048. Its repository was published on 2026-02-24. The license listed on Hugging Face is apache-2.0. Most of its layers are linear-attention layers that keep a small fixed state, so only 10 of 40 grow with the window and long contexts cost far less than on a conventional model.
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 | 35.57 GB | 0.22 GB | 35.63 GB | 99% | Lossless in practice. Use it when the memory is there. |
| Q6_K | 27.45 GB | 0.22 GB | 27.62 GB | 98% | Very close to Q8 for two thirds of the size. |
| Q5_K_M | 23.81 GB | 0.22 GB | 24.03 GB | 96% | The quality-first choice when Q6 will not fit. |
| Q4_K_M | 20.21 GB | 0.22 GB | 20.48 GB | 93% | The default. Best size-to-quality ratio for local use. |
| IQ4_XS | 17.79 GB | 0.22 GB | 18.09 GB | 91% | Importance-matrix 4-bit. Q4_K_S quality, smaller file. |
| Q3_K_M | 16.36 GB | 0.22 GB | 16.69 GB | 86% | Degradation starts to show. A way to fit one size up. |
| IQ3_XXS | 12.81 GB | 0.22 GB | 13.18 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. This model mixes 30 linear-attention layers, which keep only a small fixed state, with 10 ordinary layers, so its cache grows far more slowly than a conventional model of the same depth.
| Context | KV cache | Total VRAM | Fits in 8 GB | Fits in 12 GB | Fits in 24 GB |
|---|---|---|---|---|---|
| 4k | 0.14 GB | 20.41 GB | no | no | yes |
| 8k | 0.22 GB | 20.48 GB | no | no | yes |
| 16k | 0.37 GB | 20.64 GB | no | no | yes |
| 32k | 0.68 GB | 20.95 GB | no | no | yes |
| 64k | 1.31 GB | 21.58 GB | no | no | yes |
| 128k | 2.56 GB | 22.83 GB | no | no | yes |
The "fits" columns allow for the roughly 0.8 GB Windows keeps for the desktop.
Which hardware runs Qwen3.5 35B A3B
34 of 118 consumer devices run it at a quantization worth using, at an 8k context window.
| Device | Memory | Verdict | Quantization | Used | Speed |
|---|---|---|---|---|---|
| RTX PRO 6000 Blackwell NVIDIA | 96 GB | Runs great | Q8_0 | 35.63 GB | 251.1 t/s |
| RTX 6000 Ada Generation NVIDIA | 48 GB | Runs great | Q8_0 | 35.63 GB | 134.5 t/s |
| RTX A6000 NVIDIA | 48 GB | Runs great | Q8_0 | 35.63 GB | 107.6 t/s |
| Apple M3 Ultra 96GB Apple | 96 GB | Runs great | Q8_0 | 36.1 GB | 106.1 t/s |
| Apple M3 Ultra 256GB Apple | 256 GB | Runs great | Q8_0 | 36.1 GB | 106.1 t/s |
| Apple M3 Ultra 512GB Apple | 512 GB | Runs great | Q8_0 | 36.1 GB | 106.1 t/s |
| Apple M1 Ultra 64GB Apple | 64 GB | Runs great | Q8_0 | 36.1 GB | 103.6 t/s |
| Apple M1 Ultra 128GB Apple | 128 GB | Runs great | Q8_0 | 36.1 GB | 103.6 t/s |
| Apple M2 Ultra 64GB Apple | 64 GB | Runs great | Q8_0 | 36.1 GB | 103.6 t/s |
| Apple M2 Ultra 128GB Apple | 128 GB | Runs great | Q8_0 | 36.1 GB | 103.6 t/s |
| Apple M2 Ultra 192GB Apple | 192 GB | Runs great | Q8_0 | 36.1 GB | 103.6 t/s |
| Radeon PRO W7900 AMD | 48 GB | Runs great | Q8_0 | 35.63 GB | 102.7 t/s |
| Apple M4 Max 64GB Apple | 64 GB | Runs great | Q8_0 | 36.1 GB | 70.7 t/s |
| Apple M4 Max 128GB Apple | 128 GB | Runs great | Q8_0 | 36.1 GB | 70.7 t/s |
| Apple M1 Max 64GB Apple | 64 GB | Runs great | Q8_0 | 36.1 GB | 51.8 t/s |
| Apple M2 Max 64GB Apple | 64 GB | Runs great | Q8_0 | 36.1 GB | 51.8 t/s |
| Apple M2 Max 96GB Apple | 96 GB | Runs great | Q8_0 | 36.1 GB | 51.8 t/s |
| Apple M3 Max 64GB Apple | 64 GB | Runs great | Q8_0 | 36.1 GB | 51.8 t/s |
| Apple M3 Max 96GB Apple | 96 GB | Runs great | Q8_0 | 36.1 GB | 51.8 t/s |
| Apple M3 Max 128GB Apple | 128 GB | Runs great | Q8_0 | 36.1 GB | 51.8 t/s |
| NVIDIA DGX Spark 128GB NVIDIA | 128 GB | Runs great | Q8_0 | 36.13 GB | 38.3 t/s |
| Apple M4 Pro 64GB Apple | 64 GB | Runs great | Q8_0 | 36.1 GB | 35.4 t/s |
| Jetson AGX Orin 64GB NVIDIA | 64 GB | Runs great | Q8_0 | 36.13 GB | 28.7 t/s |
| Ryzen AI Max+ 395 64GB AMD | 64 GB | Runs great | Q8_0 | 36.13 GB | 27.7 t/s |
| Ryzen AI Max+ 395 96GB AMD | 96 GB | Runs great | Q8_0 | 36.13 GB | 27.7 t/s |
Show 35 more devices
| Ryzen AI Max+ 395 128GB AMD | 128 GB | Runs great | Q8_0 | 36.13 GB | 27.7 t/s |
| GeForce RTX 5090 NVIDIA | 32 GB | Runs great | Q6_K | 27.62 GB | 318.7 t/s |
| GeForce RTX 4090 NVIDIA | 24 GB | Runs great | Q4_K_M | 20.48 GB | 235.9 t/s |
| GeForce RTX 3090 Ti NVIDIA | 24 GB | Runs great | Q4_K_M | 20.48 GB | 235.9 t/s |
| GeForce RTX 3090 NVIDIA | 24 GB | Runs great | Q4_K_M | 20.48 GB | 219 t/s |
| GeForce RTX 5090 Laptop NVIDIA | 24 GB | Runs great | Q4_K_M | 20.48 GB | 209.6 t/s |
| Radeon RX 7900 XTX AMD | 24 GB | Runs great | Q4_K_M | 20.48 GB | 190.5 t/s |
| RTX A5000 NVIDIA | 24 GB | Runs great | Q4_K_M | 20.48 GB | 179.7 t/s |
| Radeon RX 7900 XT AMD | 20 GB | Runs great | IQ4_XS | 18.09 GB | 177.5 t/s |
| Apple M4 Max 36GB Apple | 36 GB | Just fits | n/a | 27.98 GB | 89.7 t/s |
| Apple M4 Max 48GB Apple | 48 GB | Just fits | n/a | 36.1 GB | 70.7 t/s |
| Apple M1 Max 32GB Apple | 32 GB | Just fits | n/a | 27.98 GB | 65.7 t/s |
| Apple M2 Max 32GB Apple | 32 GB | Just fits | n/a | 27.98 GB | 65.7 t/s |
| Apple M3 Max 36GB Apple | 36 GB | Just fits | n/a | 27.98 GB | 65.7 t/s |
| Apple M1 Pro 16GB Apple | 16 GB | Just fits | n/a ! | 13.33 GB | 63.9 t/s |
| Apple M2 Pro 16GB Apple | 16 GB | Just fits | n/a ! | 13.33 GB | 63.9 t/s |
| Apple M4 Pro 24GB Apple | 24 GB | Just fits | n/a | 20.74 GB | 59 t/s |
| Apple M3 Max 48GB Apple | 48 GB | Just fits | n/a | 36.1 GB | 51.8 t/s |
| Apple M5 16GB Apple | 16 GB | Just fits | n/a ! | 13.33 GB | 48.9 t/s |
| Apple M3 Pro 18GB Apple | 18 GB | Just fits | n/a ! | 13.33 GB | 47.9 t/s |
| Apple M4 16GB Apple | 16 GB | Just fits | n/a ! | 13.33 GB | 38.3 t/s |
| Jetson AGX Orin 32GB NVIDIA | 32 GB | Just fits | n/a | 28.01 GB | 36.4 t/s |
| Intel Core Ultra 9 288V 16GB Intel | 16 GB | Just fits | n/a ! | 13.36 GB | 36.3 t/s |
| Apple M4 Pro 48GB Apple | 48 GB | Just fits | n/a | 36.1 GB | 35.4 t/s |
| Ryzen AI Max+ 395 32GB AMD | 32 GB | Just fits | n/a | 28.01 GB | 35.2 t/s |
| Ryzen AI 9 HX 370 16GB AMD | 16 GB | Just fits | n/a ! | 13.36 GB | 34.2 t/s |
| Apple M5 24GB Apple | 24 GB | Just fits | n/a | 20.74 GB | 33.1 t/s |
| Apple M1 Pro 32GB Apple | 32 GB | Just fits | n/a | 27.98 GB | 32.9 t/s |
| Apple M2 Pro 32GB Apple | 32 GB | Just fits | n/a | 27.98 GB | 32.9 t/s |
| Apple M2 16GB Apple | 16 GB | Just fits | n/a ! | 13.33 GB | 31.9 t/s |
| Apple M3 16GB Apple | 16 GB | Just fits | n/a ! | 13.33 GB | 31.9 t/s |
| Apple M4 24GB Apple | 24 GB | Just fits | n/a | 20.74 GB | 25.9 t/s |
| Apple M5 32GB Apple | 32 GB | Just fits | n/a | 27.98 GB | 25.1 t/s |
| Apple M3 Pro 36GB Apple | 36 GB | Just fits | n/a | 27.98 GB | 24.6 t/s |
| Ryzen AI 9 HX 370 24GB AMD | 24 GB | Just fits | n/a | 20.77 GB | 23.1 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 |
|---|---|---|---|
| Qwen3.5 35B A3B | 20.2 GB | 55.7 t/s at Q4_K_M | 235.9 t/s at Q4_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 |
| Qwen2.5 Coder 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 Qwen3.5 35B 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 20.48 GB of its 4 GB · about 27.2 tokens/s
Best value
Radeon RX 7900 XT
$650
IQ4_XS · about 177.5 tokens/s · 800 GB/s
| Device | Price | Memory | Runs it at | Speed | |
|---|---|---|---|---|---|
| GeForce GTX 1650 NVIDIA | $100 used | 4 GB | Q4_K_M | 27.2 t/s | Check price |
| GeForce GTX 1660 SUPER NVIDIA | $110 used | 6 GB | Q4_K_M | 42.5 t/s | Check price |
| GeForce GTX 1080 Ti NVIDIA | $150 used | 11 GB | Q4_K_M | 59.7 t/s | Check price |
| GeForce RTX 2060 12GB NVIDIA | $160 used | 12 GB | Q4_K_M | 53.9 t/s | Check price |
| GeForce RTX 2080 Ti NVIDIA | $250 used | 11 GB | Q4_K_M | 64.7 t/s | Check price |
| Arc A770 16GB Intel | $280 used | 16 GB | Q4_K_M | 76.4 t/s | Check price |
| Radeon RX 6800 AMD | $330 used | 16 GB | Q4_K_M | 76.6 t/s | Check price |
| GeForce RTX 3080 12GB NVIDIA | $400 used | 12 GB | Q4_K_M | 77.2 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.
ornith-ai/Ornith-1.5-35B-A3B
Where it sits in the catalog
Source: Qwen/Qwen3.5-35B-A3B on Hugging Face. Downloaded 2.5M times in the last month. Published 2026-02-24. Architecture figures are read from the repository's own configuration file, so they move when the model does.
Qwen3 32B
32.76B · 40k context
Qwen2.5 32B Instruct
32.76B · 32k context
Qwen2.5 Coder 32B Instruct
32.76B · 32k context
Qwen3 30B A3B
30.53B (3.34B active) · 40k context
Qwen3 Coder 30B A3B Instruct
30.53B (3.34B active) · 40k context
Qwen3.5 27B
27.78B · 256k context
Qwen3 14B
14.77B · 40k context
Qwen2.5 14B Instruct
14.77B · 32k context
Questions people ask about Qwen3.5 35B A3B
How much VRAM does Qwen3.5 35B A3B need?
At Q4_K_M with an 8k context window, Qwen3.5 35B A3B needs about 20.48 GB: 19.93 GB of weights, 0.22 GB of KV cache and 0.34 GB of overhead. At the highest-quality file it needs about 35.63 GB.
What is the cheapest GPU that runs Qwen3.5 35B A3B?
The smallest consumer device in our catalog that fits it is the Radeon RX 7900 XT with 20 GB, running it at IQ4_XS at roughly 177.5 tokens per second.
How much does a longer context window add for Qwen3.5 35B A3B?
On the memory side, a 32k window adds 0.7 GB of KV cache on top of the weights, and the cache grows more slowly than a conventional model because of its attention design. The longest window the model supports natively is 256k tokens.