How much VRAM does Qwen3-0.6B need?

Q4_K_M is exactly 396,705,472 bytes (0.37 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is Qwen3-0.6B's KV cache?

3.50 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 · text

Qwen3-0.6B

Qwen/Qwen3-0.6B

Qwen3-0.6B at Q4_K_M is exactly 396,705,472 bytes (0.37 GiB / 0.40 GB) — an effective 4.222 bits per weight, not the nominal 4. Its KV cache at 32K is 3.50 GiB.

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

Parameters

752M

Architecture

qwen3

28 layers

Context

40,960

native (config.json)

License

—

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
UD-IQ1_S	0.20 GiB	214,643,392	2.285	—	unsloth
UD-IQ1_M	0.21 GiB	220,754,624	2.350	—	unsloth
UD-IQ2_XXS	0.22 GiB	234,074,816	2.491	—	unsloth
UD-IQ2_M	0.25 GiB	268,702,400	2.860	—	unsloth
UD-IQ3_XXS	0.26 GiB	282,088,128	3.002	—	unsloth
Q2_K	0.28 GiB	296,238,784	3.153	—	unsloth
Q2_K_L	0.28 GiB	296,238,784	3.153	—	unsloth
Q3_K_S	0.30 GiB	323,075,776	3.439	—	unsloth
IQ2_M	0.31 GiB	331,761,056	3.531	—	bartowski
IQ3_XXS	0.32 GiB	345,867,680	3.681	—	bartowski
Q3_K_M	0.32 GiB	347,127,488	3.695	310	unsloth
Q2_K	0.32 GiB	347,288,992	3.696	—	bartowski
IQ4_XS	0.34 GiB	367,804,096	3.915	310	unsloth
IQ3_XS	0.35 GiB	379,605,408	4.040	—	bartowski
IQ4_NL	0.36 GiB	381,566,656	4.061	—	unsloth
Q4_0	0.36 GiB	382,156,480	4.067	310	unsloth
Q4_K_S	0.36 GiB	383,270,592	4.079	—	unsloth
Q3_K_S	0.36 GiB	389,927,328	4.150	—	bartowski
Q4_K_M	0.37 GiB	396,705,472	4.222	310	unsloth
IQ3_M	0.38 GiB	402,878,880	4.288	—	bartowski
Q4_1	0.38 GiB	409,091,776	4.354	—	unsloth
Q3_K_M	0.39 GiB	413,979,040	4.406	311	bartowski
Q3_K_L	0.41 GiB	435,343,776	4.634	—	bartowski
Q5_K_S	0.41 GiB	436,616,896	4.647	—	unsloth
Q5_K_M	0.41 GiB	444,415,680	4.730	310	unsloth
IQ4_XS	0.42 GiB	450,456,992	4.794	311	bartowski
IQ4_NL	0.44 GiB	469,081,504	4.993	—	bartowski
Q4_0	0.44 GiB	469,671,328	4.999	311	bartowski
Q4_K_S	0.44 GiB	470,785,440	5.011	—	bartowski
Q4_K_M	0.45 GiB	484,220,320	5.154	311	bartowski
Q6_K	0.46 GiB	495,107,776	5.270	310	unsloth
Q2_K_L	0.46 GiB	499,224,992	5.314	—	bartowski
Q4_1	0.47 GiB	506,330,528	5.389	—	bartowski
Q5_K_S	0.51 GiB	543,579,552	5.786	—	bartowski
Q5_K_M	0.51 GiB	551,378,336	5.869	311	bartowski
Q4_K_L	0.56 GiB	599,691,680	6.383	—	bartowski
Q6_K	0.58 GiB	622,733,728	6.628	311	bartowski
Q8_0	0.60 GiB	639,446,688	6.806	310	Qwen
Q8_0	0.60 GiB	639,447,744	6.806	—	unsloth
Q5_K_L	0.60 GiB	647,401,888	6.891	—	bartowski

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.44 GiB	0.44 GiB	—	28 / 0 / 0
8,192	0.88 GiB	0.88 GiB	—	28 / 0 / 0
16,384	1.75 GiB	1.75 GiB	—	28 / 0 / 0
32,768	3.50 GiB	3.50 GiB	—	28 / 0 / 0
65,536	7.00 GiB	7.00 GiB	—	28 / 0 / 0
131,072	14.00 GiB	14.00 GiB	—	28 / 0 / 0

Compare with

same modality, comparable size

vs Llama-3.2-1B-Instruct vs ced-base vs embeddinggemma-300m vs LFM2.5-1.2B-Instruct vs MiniCPM5-1B-Claude-Opus-Fable5-Thinking vs Qwen2.5-0.5B-Instruct 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 0.39 GiB. The real file is 0.37 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

1024

Vocab

151,936

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

false

Questions people ask

How much VRAM does Qwen3-0.6B need?: Q4_K_M is exactly 396,705,472 bytes (0.37 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Qwen3-0.6B's KV cache?: 3.50 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 Qwen3-0.6B 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.