How much VRAM does ThinkingCap-Qwen3.6-27B need?

Q4_K_M is exactly 16,810,713,056 bytes (15.66 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is ThinkingCap-Qwen3.6-27B's KV cache?

2.00 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.

bottlecapai · vision language

ThinkingCap-Qwen3.6-27B

bottlecapai/ThinkingCap-Qwen3.6-27B

ThinkingCap-Qwen3.6-27B at Q4_K_M is exactly 16,810,713,056 bytes (15.66 GiB / 16.81 GB) — an effective 4.916 bits per weight, not the nominal 4. Its KV cache at 32K is 2.00 GiB.

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

Parameters

27.4B

Architecture

qwen35

64 layers

Context

262,144

native (config.json)

License

—

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
IQ2_XS	9.30 GiB	9,986,798,272	2.921	—	bartowski
IQ2_S	9.59 GiB	10,295,329,472	3.011	—	bartowski
IQ2_M	10.13 GiB	10,873,356,992	3.180	—	bartowski
Q2_K	11.03 GiB	11,839,439,552	3.462	—	bartowski
IQ3_XXS	11.76 GiB	12,626,772,672	3.692	—	bartowski
Q2_K_L	12.18 GiB	13,081,039,552	3.825	—	bartowski
IQ3_XS	12.41 GiB	13,330,404,032	3.898	—	bartowski
Q3_K_S	12.78 GiB	13,720,343,232	4.012	—	bartowski
IQ3_M	12.95 GiB	13,903,516,352	4.066	—	bartowski
Q3_K_M	13.60 GiB	14,605,734,592	4.271	866	bartowski
Q3_K_L	14.23 GiB	15,279,444,672	4.468	—	bartowski
IQ4_XS	14.50 GiB	15,567,823,552	4.553	866	bartowski
IQ4_NL	15.20 GiB	16,325,829,312	4.774	—	bartowski
Q4_0	15.23 GiB	16,348,766,912	4.781	866	bartowski
Q4_K_S	15.57 GiB	16,713,147,072	4.888	—	bartowski
Q4_K_M	15.66 GiB	16,810,713,056	4.916	866	bottlecapai
Q4_K_M	16.55 GiB	17,772,536,512	5.197	—	bartowski
Q4_1	16.60 GiB	17,825,456,832	5.213	—	bartowski
Q4_K_L	17.43 GiB	18,716,152,512	5.473	—	bartowski
Q5_K_S	18.33 GiB	19,680,944,832	5.755	—	bartowski
Q5_K_M	19.33 GiB	20,752,786,112	6.069	866	bartowski
Q5_K_L	20.06 GiB	21,537,477,312	6.298	—	bartowski
Q6_K	20.89 GiB	22,430,998,496	6.560	866	bottlecapai
Q6_K	21.85 GiB	23,463,129,792	6.861	866	bartowski
Q6_K_L	22.43 GiB	24,078,963,392	7.042	—	bartowski
Q8_0	27.05 GiB	29,047,082,976	8.494	866	bottlecapai
Q8_0	27.12 GiB	29,116,388,032	8.515	866	bartowski
F16	50.90 GiB	54,657,732,576	15.984	866	bottlecapai

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.25 GiB	1.00 GiB	4.00×	16 / 0 / 48
8,192	0.50 GiB	2.00 GiB	4.00×	16 / 0 / 48
16,384	1.00 GiB	4.00 GiB	4.00×	16 / 0 / 48
32,768	2.00 GiB	8.00 GiB	4.00×	16 / 0 / 48
65,536	4.00 GiB	16.00 GiB	4.00×	16 / 0 / 48
131,072	8.00 GiB	32.00 GiB	4.00×	16 / 0 / 48

48 of 64 layers use linear attention, which keeps a fixed-size recurrent state instead of a per-token cache. Those layers do not grow with context at all — treating them as ordinary attention, as a flat formula does, overstates this model's cache by roughly 4.0× at long context.

Compare with

same modality, comparable size

vs Qwen3.6-35B-A3B vs Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP vs Qwen3.5-9B vs gemma-4-26B-A4B-it vs gemma-4-12B-it vs Qwen3.5-4B 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 14.33 GiB. The real file is 15.66 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

256

Hidden size

5120

Vocab

248,320

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does ThinkingCap-Qwen3.6-27B need?: Q4_K_M is exactly 16,810,713,056 bytes (15.66 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is ThinkingCap-Qwen3.6-27B's KV cache?: 2.00 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.
Which quantization of ThinkingCap-Qwen3.6-27B 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.