OpenGVLab · vision language

InternVL3_5-30B-A3B

OpenGVLab/InternVL3_5-30B-A3B

InternVL3_5-30B-A3B at Q4_K_M is exactly 18,632,180,352 bytes (17.35 GiB / 18.63 GB) — an effective 4.832 bits per weight, not the nominal 4.

From the file· summed from 1 file(s)From the file· KV per layer
Parameters
30.8B
Architecture
qwen3moe
null layers
Context
—
native (config.json)
License
—

Shipped quantizations

● exact bytes, summed from published files
QuantSize●Exact bytes●Effective bpw●Tensors●Publisher
IQ2_XXS7.05 GiB7,569,021,8561.963—bartowski
IQ2_XS8.07 GiB8,662,162,3362.246—bartowski
IQ2_S8.14 GiB8,744,096,6722.268—bartowski
IQ2_M9.19 GiB9,870,267,2962.560—bartowski
Q2_K10.16 GiB10,908,644,2562.829—bartowski
Q2_K_L10.44 GiB11,212,516,2562.908—bartowski
IQ3_XXS11.38 GiB12,216,980,3843.168—bartowski
IQ3_XS11.86 GiB12,736,457,6323.303—bartowski
Q3_K_S12.51 GiB13,428,124,5763.482—bartowski
IQ3_M13.11 GiB14,076,537,7603.651—bartowski
Q3_K_M13.11 GiB14,076,799,9043.651—bartowski
Q3_K_L13.58 GiB14,582,999,6803.782—lmstudio-community
Q3_K_L13.58 GiB14,582,999,9683.782—bartowski
IQ4_XS15.33 GiB16,457,999,2644.268—bartowski
IQ4_NL16.19 GiB17,386,275,7444.509—bartowski
Q4_016.42 GiB17,631,642,5284.572—bartowski
Q4_K_S16.75 GiB17,984,488,3524.664—bartowski
Q4_K_M17.35 GiB18,632,180,3524.832—lmstudio-community
Q4_K_M17.35 GiB18,632,180,6404.832—bartowski
Q4_K_L17.57 GiB18,863,123,3604.892—bartowski
Q4_117.89 GiB19,214,517,1524.983—bartowski
Q5_K_S19.65 GiB21,099,381,6645.472—bartowski
Q5_K_M20.25 GiB21,744,452,5125.639—bartowski
Q5_K_L20.43 GiB21,936,499,6165.689—bartowski
Q6_K23.38 GiB25,104,718,4646.510—lmstudio-community
Q6_K23.38 GiB25,104,718,7526.510—bartowski
Q6_K_L23.52 GiB25,255,439,2646.550—bartowski
Q8_030.25 GiB32,483,928,7048.424—lmstudio-community
Q8_030.25 GiB32,483,928,9928.424—bartowski
BF162 shards56.90 GiB61,095,799,64815.844—bartowski

Compare with

same modality, comparable size
vs Qwen3.6-35B-A3Bvs Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTPvs Qwen3.5-9Bvs gemma-4-26B-A4B-itvs gemma-4-12B-itvs Qwen3.5-4BAll 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 16.16 GiB. The real file is 17.35 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
—
Hidden size
—
Vocab
—
Sliding window
none
SWA period
—
MLA
no
Experts
—
Experts per token
—
use_sliding_window
—

Questions people ask

How much VRAM does InternVL3_5-30B-A3B need?
Q4_K_M is exactly 18,632,180,352 bytes (17.35 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
Which quantization of InternVL3_5-30B-A3B 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.