How much VRAM does whisper-medium need?

Q4_K_M is exactly 504,102,848 bytes (0.47 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

openai · audio asr

whisper-medium

openai/whisper-medium

whisper-medium at Q4_K_M is exactly 504,102,848 bytes (0.47 GiB / 0.50 GB) — an effective 5.279 bits per weight, not the nominal 4.

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

Parameters

764M

Architecture

whisper

24 layers

Context

—

native (config.json)

License

apache-2.0

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
Q2_K	0.25 GiB	266,895,060	2.795	—	oxide-lab
Q3_K	0.32 GiB	343,854,332	3.601	—	oxide-lab
Q4_0	0.40 GiB	434,685,376	4.553	—	oxide-lab
Q4_0	0.41 GiB	444,493,380	4.655	—	oxide-lab
Q4_K	0.41 GiB	444,493,380	4.655	—	oxide-lab
Q4_1	0.45 GiB	482,214,528	5.050	—	oxide-lab
Q4_1	0.46 GiB	491,852,932	5.151	—	oxide-lab
Q4_K_M	0.47 GiB	504,102,848	5.279	949	handy-computer
Q5_0	0.50 GiB	539,212,484	5.647	—	oxide-lab
Q5_K	0.50 GiB	539,212,484	5.647	—	oxide-lab
Q5_K_M	0.54 GiB	582,746,048	6.103	949	handy-computer
Q5_1	0.55 GiB	586,572,036	6.143	—	oxide-lab
Q6_K	0.60 GiB	639,851,532	6.701	—	oxide-lab
Q6_K	0.60 GiB	648,019,904	6.787	949	handy-computer
Q8_0	0.76 GiB	814,918,592	8.535	948	oxide-lab
Q8_0	0.77 GiB	823,369,796	8.623	—	oxide-lab
Q8_0	0.77 GiB	831,538,144	8.709	949	handy-computer
F16	1.44 GiB	1,541,931,424	16.149	949	handy-computer
F32	2.85 GiB	3,057,437,088	32.021	—	handy-computer

Compare with

same modality, comparable size

vs nemotron-3.5-asr-streaming-0.6b vs parakeet-unified-en-0.6b vs parakeet-tdt-0.6b-v3 vs whisper-large-v3 vs canary-180m-flash vs whisper-large-v3-turbo 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 0.40 GiB. The real file is 0.47 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

—

Hidden size

—

Vocab

51,865

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does whisper-medium need?: Q4_K_M is exactly 504,102,848 bytes (0.47 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
Which quantization of whisper-medium 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.