How much VRAM does GLM-OCR need?

Q4_K_M is exactly 548,515,136 bytes (0.51 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is GLM-OCR'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.

zai-org · text

GLM-OCR

zai-org/GLM-OCR

GLM-OCR at Q4_K_M is exactly 548,515,136 bytes (0.51 GiB / 0.55 GB) — an effective 3.311 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

1.3B

Architecture

glm4

16 layers

Context

131,072

native (config.json)

License

—

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
Q2_K	0.34 GiB	361,356,608	2.181	—	mradermacher
I1-Q2_K	0.34 GiB	361,356,832	2.181	—	mradermacher
I1-IQ3_XS	0.38 GiB	410,496,544	2.478	—	mradermacher
Q3_K_S	0.39 GiB	422,219,072	2.549	—	mradermacher
I1-Q3_K_S	0.39 GiB	422,219,296	2.549	—	mradermacher
I1-IQ3_S	0.39 GiB	422,219,296	2.549	—	mradermacher
I1-IQ3_M	0.40 GiB	434,753,056	2.624	—	mradermacher
Q3_K_M	0.42 GiB	450,131,264	2.717	—	mradermacher
I1-Q3_K_M	0.42 GiB	450,131,488	2.717	—	mradermacher
Q3_K_L	0.44 GiB	473,920,832	2.861	—	mradermacher
I1-Q3_K_L	0.44 GiB	473,921,056	2.861	—	mradermacher
I1-IQ4_XS	0.47 GiB	503,449,120	3.039	—	mradermacher
IQ4_XS	0.47 GiB	505,660,736	3.053	—	mradermacher
I1-IQ4_NL	0.49 GiB	528,442,912	3.190	—	mradermacher
I1-Q4_0	0.49 GiB	529,327,648	3.195	—	mradermacher
Q4_K_S	0.49 GiB	530,998,592	3.205	—	mradermacher
I1-Q4_K_S	0.49 GiB	530,998,816	3.205	—	mradermacher
Q4_K_M	0.51 GiB	548,515,136	3.311	—	mradermacher
I1-Q4_K_M	0.51 GiB	548,515,360	3.311	—	mradermacher
I1-Q4_1	0.54 GiB	578,430,496	3.492	—	mradermacher
Q5_K_S	0.59 GiB	628,417,856	3.793	—	mradermacher
I1-Q5_K_S	0.59 GiB	628,418,080	3.793	—	mradermacher
Q5_K_M	0.59 GiB	638,758,208	3.856	—	mradermacher
I1-Q5_K_M	0.59 GiB	638,758,432	3.856	—	mradermacher
Q6_K	0.68 GiB	734,641,472	4.435	—	mradermacher
I1-Q6_K	0.68 GiB	734,641,696	4.435	—	mradermacher
Q8_0	0.89 GiB	950,433,408	5.737	—	ggml-org
Q8_0	0.89 GiB	950,437,184	5.737	—	mradermacher
F16	1.66 GiB	1,785,771,648	10.780	—	ggml-org
F16	1.66 GiB	1,785,775,424	10.780	—	mradermacher

KV cache by context

computed per layer

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

Compare with

same modality, comparable size

vs Llama-3.2-1B-Instruct vs ced-base vs gemma-3-1b-it vs Qwen3-1.7B vs embeddinggemma-300m vs Qwen3-0.6B 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 0.69 GiB. The real file is 0.51 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

1536

Vocab

59,392

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does GLM-OCR need?: Q4_K_M is exactly 548,515,136 bytes (0.51 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-OCR'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 GLM-OCR 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.