How much VRAM does Hunyuan-A13B-Instruct need?

Q4_K_M is exactly 48,782,258,016 bytes (45.43 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is Hunyuan-A13B-Instruct's KV cache?

4.00 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.

Is Hunyuan-A13B-Instruct a mixture-of-experts model?

Yes — 64 experts with null routed per token. Every expert must be resident in memory, but only the routed ones are read per token, so its memory requirement and its speed behave very differently.

tencent · text · mixture of experts

Hunyuan-A13B-Instruct

tencent/Hunyuan-A13B-Instruct

Hunyuan-A13B-Instruct at Q4_K_M is exactly 48,782,258,016 bytes (45.43 GiB / 48.78 GB) — an effective 4.854 bits per weight, not the nominal 4. Its KV cache at 32K is 4.00 GiB.

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

Parameters

80.4B

total, not active

Architecture

hunyuan-moe

32 layers

Context

32,768

native (config.json)

License

other

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
IQ1_S	15.18 GiB	16,298,101,888	1.622	—	gabriellarson
IQ1_M	16.54 GiB	17,754,901,760	1.767	—	bartowski
IQ1_M	16.87 GiB	18,115,742,848	1.803	—	gabriellarson
IQ2_XXS	18.39 GiB	19,743,526,144	1.965	—	bartowski
IQ2_XXS	19.69 GiB	21,145,144,448	2.104	—	gabriellarson
UD-TQ1_0	20.95 GiB	22,495,637,888	2.239	—	unsloth
IQ2_XS	21.11 GiB	22,666,956,032	2.256	—	bartowski
IQ2_S	21.17 GiB	22,734,064,896	2.262	—	bartowski
IQ2_XS	21.97 GiB	23,585,442,944	2.347	—	gabriellarson
IQ2_S	22.19 GiB	23,827,074,176	2.371	—	gabriellarson
UD-IQ1_S	22.89 GiB	24,578,109,824	2.446	—	unsloth
UD-IQ1_M	23.82 GiB	25,572,651,392	2.545	—	unsloth
IQ2_M	23.95 GiB	25,714,117,888	2.559	—	bartowski
IQ2_M	24.45 GiB	26,250,595,456	2.612	—	gabriellarson
Q2_K_S	25.59 GiB	27,472,310,112	2.734	—	gabriellarson
UD-IQ2_XXS	25.73 GiB	27,625,271,680	2.749	—	unsloth
UD-IQ2_M	26.85 GiB	28,833,231,232	2.869	—	unsloth
Q2_K	26.98 GiB	28,966,268,896	2.882	—	bartowski
Q2_K_L	27.10 GiB	29,093,410,560	2.895	—	bartowski
Q2_K	27.40 GiB	29,418,205,024	2.927	—	gabriellarson
Q2_K_L	27.40 GiB	29,418,205,568	2.927	—	unsloth
Q2_K	27.40 GiB	29,418,205,568	2.927	—	unsloth
IQ3_XXS	28.85 GiB	30,979,017,856	3.083	—	gabriellarson
IQ3_XXS	30.19 GiB	32,415,567,104	3.226	—	bartowski
IQ3_XS	30.67 GiB	32,930,934,624	3.277	—	gabriellarson
IQ3_XS	31.41 GiB	33,726,803,936	3.356	—	bartowski
UD-IQ3_XXS	31.88 GiB	34,229,694,848	3.406	—	unsloth
Q3_K_S	32.40 GiB	34,784,816,992	3.462	—	gabriellarson
Q3_K_S	32.40 GiB	34,784,817,536	3.462	—	unsloth
IQ3_S	32.41 GiB	34,802,642,784	3.463	—	gabriellarson
IQ3_M	32.88 GiB	35,308,449,632	3.514	—	gabriellarson
Q3_K_S	33.10 GiB	35,546,083,296	3.537	—	bartowski
IQ3_M	34.72 GiB	37,275,185,120	3.709	—	bartowski
Q3_K_M	34.72 GiB	37,276,233,696	3.709	—	bartowski
Q3_K_M	35.91 GiB	38,555,496,288	3.837	—	gabriellarson
Q3_K_M	35.91 GiB	38,555,496,832	3.837	—	unsloth
Q3_K_L	36.10 GiB	38,764,163,040	3.857	—	bartowski
Q3_K_L	38.84 GiB	41,705,418,592	4.150	—	gabriellarson
IQ4_XS	39.97 GiB	42,916,523,872	4.271	—	gabriellarson
IQ4_XS	39.97 GiB	42,916,524,416	4.271	—	unsloth

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.50 GiB	0.50 GiB	—	32 / 0 / 0
8,192	1.00 GiB	1.00 GiB	—	32 / 0 / 0
16,384	2.00 GiB	2.00 GiB	—	32 / 0 / 0
32,768	4.00 GiB	4.00 GiB	—	32 / 0 / 0
65,536	8.00 GiB	8.00 GiB	—	32 / 0 / 0
131,072	16.00 GiB	16.00 GiB	—	32 / 0 / 0

Compare with

same modality, comparable size

vs Qwen3-Coder-30B-A3B-Instruct vs Qwen3.6-27B vs Qwen3.8-27B vs gemma-4-12B-it-qat-q4_0-unquantized vs gemma-4-E4B-it vs Qwen3-30B-A3B-Thinking-2507 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 42.12 GiB. The real file is 45.43 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

4096

Vocab

128,167

Sliding window

none

SWA period

—

MLA

Experts

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does Hunyuan-A13B-Instruct need?: Q4_K_M is exactly 48,782,258,016 bytes (45.43 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Hunyuan-A13B-Instruct's KV cache?: 4.00 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.
Is Hunyuan-A13B-Instruct a mixture-of-experts model?: Yes — 64 experts, null routed per token. Every expert must be resident, but only the routed ones are read per token, which is why its memory requirement and its speed behave very differently.
Which quantization of Hunyuan-A13B-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.