How much VRAM does Harbinger-24B need?

Q4_K_M is exactly 14,333,907,904 bytes (13.35 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is Harbinger-24B's KV cache?

5.00 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.

LatitudeGames · text

Harbinger-24B

LatitudeGames/Harbinger-24B

Harbinger-24B at Q4_K_M is exactly 14,333,907,904 bytes (13.35 GiB / 14.33 GB) — an effective 4.865 bits per weight, not the nominal 4. Its KV cache at 32K is 5.00 GiB.

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

Parameters

23.6B

Architecture

llama

40 layers

Context

131,072

native (config.json)

License

apache-2.0

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
IQ1_S	4.91 GiB	5,273,719,744	1.790	—	LatitudeGames
IQ1_M	5.36 GiB	5,750,494,144	1.952	—	LatitudeGames
IQ2_XXS	6.10 GiB	6,545,118,144	2.221	—	LatitudeGames
IQ2_XS	6.71 GiB	7,207,031,744	2.446	—	LatitudeGames
IQ2_S	6.96 GiB	7,478,350,784	2.538	—	LatitudeGames
IQ2_M	7.56 GiB	8,114,049,984	2.754	—	LatitudeGames
Q2_K_S	7.75 GiB	8,320,160,704	2.824	—	LatitudeGames
Q2_K	8.28 GiB	8,890,323,904	3.017	—	LatitudeGames
IQ3_XXS	8.64 GiB	9,280,590,784	3.150	—	LatitudeGames
IQ3_XS	9.23 GiB	9,907,114,944	3.362	—	LatitudeGames
Q3_K_S	9.69 GiB	10,400,273,344	3.530	—	LatitudeGames
IQ3_S	9.71 GiB	10,428,126,144	3.539	—	LatitudeGames
IQ3_M	9.92 GiB	10,650,948,544	3.615	—	LatitudeGames
Q3_K_M	10.69 GiB	11,474,080,704	3.894	—	LatitudeGames
Q3_K_L	11.55 GiB	12,400,759,744	4.209	—	LatitudeGames
IQ4_XS	11.88 GiB	12,758,913,984	4.330	—	LatitudeGames
IQ4_NL	12.54 GiB	13,468,013,504	4.571	—	LatitudeGames
Q4_K_S	12.62 GiB	13,549,278,144	4.598	—	LatitudeGames
Q4_K_M	13.35 GiB	14,333,907,904	4.865	—	LatitudeGames
Q5_K_S	15.18 GiB	16,304,411,584	5.533	—	LatitudeGames
Q5_K_M	15.61 GiB	16,763,982,784	5.689	—	LatitudeGames
Q6_K	18.02 GiB	19,345,937,344	6.566	—	LatitudeGames
Q8_0	23.33 GiB	25,054,778,304	8.503	—	LatitudeGames
BF16	43.92 GiB	47,153,517,280	16.003	—	LatitudeGames

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.63 GiB	0.63 GiB	—	40 / 0 / 0
8,192	1.25 GiB	1.25 GiB	—	40 / 0 / 0
16,384	2.50 GiB	2.50 GiB	—	40 / 0 / 0
32,768	5.00 GiB	5.00 GiB	—	40 / 0 / 0
65,536	10.00 GiB	10.00 GiB	—	40 / 0 / 0
131,072	20.00 GiB	20.00 GiB	—	40 / 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 12.35 GiB. The real file is 13.35 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

5120

Vocab

131,072

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does Harbinger-24B need?: Q4_K_M is exactly 14,333,907,904 bytes (13.35 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Harbinger-24B's KV cache?: 5.00 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 Harbinger-24B 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.