zai-org · text

GLM-4.6V-Flash

zai-org/GLM-4.6V-Flash

GLM-4.6V-Flash at Q4_K_M is exactly 6,166,577,888 bytes (5.74 GiB / 6.17 GB) — an effective 4.793 bits per weight, not the nominal 4. Its KV cache at 32K is 1.25 GiB.

From the file· summed from 1 file(s)From the file· KV per layer
Parameters
10.3B
Architecture
glm4
40 layers
Context
131,072
native (config.json)
License
mit

Shipped quantizations

● exact bytes, summed from published files
QuantSize●Exact bytes●Effective bpw●Tensors●Publisher
UD-IQ1_S2.98 GiB3,204,204,1282.490—unsloth
UD-IQ1_M3.09 GiB3,318,154,8482.579—unsloth
UD-IQ2_XXS3.27 GiB3,511,158,3682.729—unsloth
IQ2_S3.52 GiB3,783,787,6482.941—bartowski
IQ2_M3.68 GiB3,946,316,9283.067—bartowski
UD-IQ2_M3.72 GiB3,992,028,7683.103—unsloth
Q2_K3.73 GiB4,006,347,9043.114—bartowski
Q2_K3.73 GiB4,006,348,3843.114—unsloth
Q2_K_L3.87 GiB4,151,838,3043.227—unsloth
IQ3_XXS3.94 GiB4,226,811,0083.285—bartowski
UD-IQ3_XXS3.94 GiB4,234,839,6483.292—unsloth
IQ3_XS4.10 GiB4,406,412,4163.425—bartowski
Q3_K_S4.28 GiB4,592,043,1363.569—bartowski
Q3_K_S4.28 GiB4,592,043,6163.569—unsloth
Q2_K_L4.30 GiB4,612,555,9043.585—bartowski
IQ3_M4.40 GiB4,721,804,4163.670—bartowski
Q3_K_M4.63 GiB4,974,511,2323.866523bartowski
Q3_K_M4.63 GiB4,974,511,7123.866523unsloth
Q3_K_L4.84 GiB5,196,612,7364.039—bartowski
IQ4_XS4.90 GiB5,262,279,8084.090523bartowski
IQ4_XS4.92 GiB5,281,678,9444.105523unsloth
IQ4_NL5.09 GiB5,465,179,2644.248—bartowski
IQ4_NL5.09 GiB5,465,179,7444.248—unsloth
Q4_05.10 GiB5,477,467,2644.257523bartowski
Q4_05.10 GiB5,477,467,7444.257523unsloth
Q4_K_S5.36 GiB5,758,485,6324.476—bartowski
Q4_K_S5.36 GiB5,758,486,1124.476—unsloth
Q4_15.60 GiB6,008,603,7764.670—bartowski
Q4_15.60 GiB6,008,604,2564.670—unsloth
Q4_K_M5.74 GiB6,166,577,8884.793—lmstudio-community
Q4_K_M5.74 GiB6,166,578,3044.793523bartowski
Q4_K_M5.74 GiB6,166,578,7844.793523unsloth
Q4_K_L6.17 GiB6,627,296,3845.151—bartowski
Q5_K_S6.24 GiB6,697,518,2085.206—bartowski
Q5_K_S6.24 GiB6,697,518,6885.206—unsloth
Q5_K_M6.57 GiB7,050,921,0885.480523bartowski
Q5_K_M6.57 GiB7,050,921,5685.480—unsloth
Q5_K_L6.92 GiB7,434,044,5445.778—bartowski
Q6_K7.70 GiB8,266,646,2406.425—lmstudio-community
Q6_K7.70 GiB8,266,646,6566.425523bartowski

KV cache by context

computed per layer
ContextKV cache (f16)●Flat formulaOverstated byFull / windowed / recurrent
4,0960.16 GiB0.16 GiB—40 / 0 / 0
8,1920.31 GiB0.31 GiB—40 / 0 / 0
16,3840.63 GiB0.63 GiB—40 / 0 / 0
32,7681.25 GiB1.25 GiB—40 / 0 / 0
65,5362.50 GiB2.50 GiB—40 / 0 / 0
131,0725.00 GiB5.00 GiB—40 / 0 / 0

Compare with

same modality, comparable size
vs Qwen3-Coder-30B-A3B-Instructvs gemma-4-12B-it-qat-q4_0-unquantizedvs gemma-4-E4B-itvs Qwen3-30B-A3B-Thinking-2507vs Qwen3-4Bvs Qwen3-8BAll alternatives →Will it fit on my card? →

Will it run on your card?

full quant x context sweep
By VRAM:8GB12GB16GB24GB32GB48GB
Radeon RX 6500 XT 4GBGeForce RTX 3050 6GBGeForce RTX 5050 8GBGeForce RTX 3080 10GBGeForce RTX 2080 Ti 11GBGeForce RTX 5070 12GBGeForce RTX 5060 Ti 16GBApple M3 Pro 18GBGeForce RTX 3080 Ti 20GBGeForce RTX 5090 D V2 24GBGeForce RTX 5090 D 32GBApple M5 Max 36GBApple M5 Pro 48GBApple M5 Max 64GBApple M3 Ultra 96GBApple M5 Max 128GBApple M2 Ultra 192GBApple M3 Ultra 256GBApple M3 Ultra 512GB

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 5.39 GiB. The real file is 5.74 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

from config.json
Layers
40
Attention heads
32
KV heads
2
Head dim
128
Hidden size
4096
Vocab
151,552
Sliding window
none
SWA period
—
MLA
no
Experts
—
Experts per token
—
use_sliding_window
—

Questions people ask

How much VRAM does GLM-4.6V-Flash need?
Q4_K_M is exactly 6,166,577,888 bytes (5.74 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is GLM-4.6V-Flash's KV cache?
1.25 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 GLM-4.6V-Flash 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.