IQ3_S GGUF models

A GGUF quantization format. Lower numbers mean smaller files and more quality loss. 45340 models have a IQ3_S file. Sizes run from 1M to 1030B parameters. The longest trained context is 102400K tokens.

Most downloadedNewestLargestLongest contextSmallest
QW
Qwen3.8 Flash Next GSQ RCO
Qwen · ISTA-DASLab
Params177B
Context256K
Smallest66.4 GB
Vision
QW
Qwen3.8-27B
Qwen · unsloth
Params27.3B
Context256K
Q4_K_M16.5 GB
Vision
QW
Qwen3.8 Flash Next GSQ RCO abliterated
Qwen · SC117
Params177B
Context256K
Smallest0.9 GB
GGUF
QW
Qwen3.8-27B
Qwen · huihui-ai
Params27.3B
Context256K
Q4_K_M16.6 GB
Vision
OR
Tiel Coder 35B A3B GGUF MTP
Ornith · peculiar-ragdoll
Params447M
Context256K
Smallest12.7 GB
CodingVision
QW
Qwen3.8 27B
Qwen · byteshape
Params27.3B
Context256K
Smallest8.8 GB
Vision
QW
Swift 1.5 Qwen3.8 27B GSQ RCO
Qwen · ukisai
Params27B
Context—
Smallest0.0 GB
Reasoning
QW
Qwen3.8-27B
Qwen · peculiar-ragdoll
Params27.3B
Context256K
Q4_K_M16.5 GB
Vision
QW
Qwen3.6-35B-A3B
Qwen · unsloth
Params34.7B
Context256K
Q4_K_M22.1 GB
Vision
QW
Qwen3.6-35B-A3B
Qwen · unsloth
Params35.5B
Context256K
Q4_K_M22.7 GB
Vision
QW
Qwen3.8 27B GSQ RCO
Qwen · ISTA-DASLab
Params27B
Context—
Smallest0.0 GB
Vision
WA
Wan2.1
Wan · DeepBeepMeep
Params9.2B
Context256K
Q4_K_M0.3 GB
GGUF
GE
Gemma-4-26B-A4B-It
Gemma · unsloth
Params25.2B
Context256K
Q4_K_M16.9 GB
Vision
QW
Qwen3.5-35B-A3B
Qwen · unsloth
Params34.7B
Context256K
Q4_K_M22.0 GB
Vision
HU
Cyber Tiel Coder 35B A3B GGUF MTP
Huihui · peculiar-ragdoll
Params35.5B
Context256K
Q4_K_M22.5 GB
CodingReasoningTool use
QW
Qwen3.8-27B
Qwen · agentionai
Params27.3B
Context256K
Q4_K_M16.5 GB
ReasoningTool useVision
CY
Cyber Ornith 1.5 9B OBLITERATED
Cyber · mradermacher
Params8.95B
Context256K
Q4_K_M5.6 GB
CodingReasoningTool use
QW
Swift 1.5 Qwen3.8 Flash Next GSQ RCO
Qwen · ukisai
Params177B
Context—
Smallest0.6 GB
ReasoningVision
IN
Inkling-Small
Inkling · unsloth
Params264B
Context1024K
Q4_K_M162.5 GB
ReasoningVision
QW
Swift 1.5 Qwen3.8 27B heretic GSQ RCO
Qwen · jegreene
Params27.3B
Context256K
Smallest9.6 GB
ReasoningTool useVision
QW
Qwen3.8 27B
Qwen · DavidAU
Params27B
Context256K
Q4_K_M16.5 GB
CodingReasoningVision
GL
GLM 5.3 Flash Maya
GLM · peasantsmith
Params370B
Context64K
Smallest0.0 GB
Vision
MO
models moved
Models · ggml-org
Params34M
Context—
Smallest0.0 GB
GGUF
QW
RVN Qwen3.8 Flash Next Abliterated Uncensored IQ3 S Strata
Qwen · jimmy31chen
Params177B
Context256K
Smallest88.9 GB
GGUF
QW
Swift 1.5 Qwen3.8 Flash Next GSQ RCO abliterated
Qwen · SC117
Params177B
Context256K
Smallest0.9 GB
GGUF
OR
Tiel Coder 35B A3B
Ornith · peculiar-ragdoll
Params34.7B
Context256K
Smallest0.2 GB
GGUF
QW
Qwen3.8 Flash Next abliterated GSQ RCO Strata
Qwen · alesha-pro
Params—
Context—
Smallest0.0 GB
GGUF
QW
Huihui Qwen3.8 Flash Next abliterated v2
Qwen · huihui-ai
Params177B
Context—
Smallest1.2 GB
ReasoningVision
QW
Qwen-Agentworld-35B-A3B
Qwen · unsloth
Params34.7B
Context256K
Q4_K_M22.1 GB
GGUF
LI
Ling 3.0 Tiny
Ling · bloomer010
Params7.89B
Context128K
Q4_K_M4.8 GB
Coding

What is IQ3_S?

A GGUF quantization format. Lower numbers mean smaller files and more quality loss. On average it stores about 3.5 bits per weight, so a IQ3_S file is roughly 22% of the size of the same model in 16-bit. A reasonable choice when a 4-bit file will not fit, mostly on larger models. Quality loss is noticeable but often acceptable.

45,340 models have a IQ3_S file on ggufmodels.com, and those files range from 0.0 to 443.6 GB. The most downloaded are Qwen3.8 Flash Next GSQ RCO, Qwen3.8-27B and Qwen3.8 Flash Next GSQ RCO abliterated. For a longer explanation of every quant type, read GGUF quantization explained.

Frequently asked questions

What is IQ3_S in GGUF?

A GGUF quantization format. Lower numbers mean smaller files and more quality loss. It averages about 3.5 bits per weight.

Should I use IQ3_S?

A reasonable choice when a 4-bit file will not fit, mostly on larger models. Quality loss is noticeable but often acceptable.

How big is a IQ3_S file?

IQ3_S files on this site range from 0.0 to 443.6 GB, depending on the model. As a rule of thumb it is about 22% of the 16-bit size.

Other quants

Keep browsing