How much VRAM does Qwen3-VL-4B-Instruct need?

Q4_K_M is exactly 2,497,281,568 bytes (2.33 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is Qwen3-VL-4B-Instruct's KV cache?

4.50 GiB at 32K context with an f16 cache, computed per layer from the model's architecture. Quantizing the cache to q8_0 roughly halves it.

Qwen · vision language

Qwen3-VL-4B-Instruct

Qwen/Qwen3-VL-4B-Instruct

Qwen3-VL-4B-Instruct at Q4_K_M is exactly 2,497,281,568 bytes (2.33 GiB / 2.50 GB) — an effective 4.502 bits per weight, not the nominal 4. Its KV cache at 32K is 4.50 GiB.

From the file· summed from 1 file(s)From the file· KV per layer

Parameters

4.4B

Architecture

qwen3vl

36 layers

Context

262,144

native (config.json)

License

apache-2.0

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
UD-IQ1_S	1.01 GiB	1,083,151,136	1.953	—	unsloth
UD-IQ1_M	1.06 GiB	1,143,485,216	2.061	—	unsloth
UD-IQ2_XXS	1.17 GiB	1,255,346,976	2.263	—	unsloth
UD-IQ2_M	1.43 GiB	1,530,403,616	2.759	—	unsloth
Q2_K	1.55 GiB	1,669,501,216	3.010	—	unsloth
Q2_K_L	1.55 GiB	1,669,501,216	3.010	—	unsloth
UD-IQ3_XXS	1.56 GiB	1,674,378,016	3.018	—	unsloth
Q3_K_S	1.76 GiB	1,886,998,816	3.402	—	unsloth
Q3_K_M	1.93 GiB	2,075,619,616	3.742	398	unsloth
IQ4_XS	2.11 GiB	2,270,753,056	4.093	398	unsloth
Q4_0	2.21 GiB	2,375,774,496	4.283	398	unsloth
IQ4_NL	2.22 GiB	2,381,345,056	4.293	—	unsloth
Q4_K_S	2.22 GiB	2,383,311,136	4.296	—	unsloth
Q4_K_M	2.33 GiB	2,497,281,568	4.502	—	lmstudio-community
Q4_K_M	2.33 GiB	2,497,281,664	4.502	398	Qwen
Q4_K_M	2.33 GiB	2,497,282,336	4.502	—	unsloth
Q4_1	2.42 GiB	2,596,630,816	4.681	—	unsloth
Q5_K_S	2.63 GiB	2,823,713,056	5.090	—	unsloth
Q5_K_M	2.69 GiB	2,889,515,296	5.209	398	unsloth
Q6_K	3.08 GiB	3,306,262,048	5.960	398	lmstudio-community
Q6_K	3.08 GiB	3,306,262,816	5.960	398	unsloth
Q8_0	3.99 GiB	4,280,406,048	7.716	—	lmstudio-community
Q8_0	3.99 GiB	4,280,406,144	7.716	398	Qwen
Q8_0	3.99 GiB	4,280,406,816	7.716	—	unsloth
F16	7.50 GiB	8,051,286,144	14.514	398	Qwen
BF16	7.50 GiB	8,051,286,560	14.514	—	unsloth

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.56 GiB	0.56 GiB	—	36 / 0 / 0
8,192	1.13 GiB	1.13 GiB	—	36 / 0 / 0
16,384	2.25 GiB	2.25 GiB	—	36 / 0 / 0
32,768	4.50 GiB	4.50 GiB	—	36 / 0 / 0
65,536	9.00 GiB	9.00 GiB	—	36 / 0 / 0
131,072	18.00 GiB	18.00 GiB	—	36 / 0 / 0

Compare with

same modality, comparable size

vs Qwen3.5-4B vs Qwen3.5-0.8B vs gemma-4-E2B-it vs Qwen2.5-VL-7B-Instruct vs Qwen3-VL-2B-Instruct vs Qwen3.5-2B All alternatives →Will it fit on my card? →

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 2.32 GiB. The real file is 2.33 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

from config.json

Layers

Attention heads

KV heads

Head dim

128

Hidden size

2560

Vocab

151,936

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does Qwen3-VL-4B-Instruct need?: Q4_K_M is exactly 2,497,281,568 bytes (2.33 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Qwen3-VL-4B-Instruct's KV cache?: 4.50 GiB at 32K context with an f16 cache, computed per layer. Quantizing the cache to q8_0 roughly halves it, which is often the difference between a context length fitting and not.
Which quantization of Qwen3-VL-4B-Instruct should I use?: Q4_K_M is the usual default. Pick the largest quantization that fits your card at the context you actually need — the table above gives exact sizes for every one published.