Gemma 3 4B Instruct quants compared

26 GGUF builds by real file size, probed from unsloth/gemma-3-4b-it-GGUF on Hugging Face (2026-09-02). 4B params.

Download Gemma 3 4B Instruct Q4_K_M (2.32 GB) — it fits 8 GB of memory with 16k context. With Ollama: ollama run gemma3:4b

Every Gemma 3 4B Instruct quant by real file size

QuantWeights+ KV (16k)TotalFits comfortably inQuality
BF167.23 GB1.8 GB9.0 GB12 GBFull precision (lossless)
Q8_K_XL4.81 GB1.8 GB6.6 GB8 GBNear-lossless
Q8_03.85 GB1.8 GB5.6 GB8 GBNear-lossless
Q6_K_XL3.32 GB1.8 GB5.1 GB8 GBExcellent
Q6_K2.97 GB1.8 GB4.7 GB8 GBExcellent
Q5_K_M2.64 GB1.8 GB4.4 GB8 GBVery high
Q5_K_XL2.64 GB1.8 GB4.4 GB8 GBVery high
Q5_K_S2.57 GB1.8 GB4.3 GB8 GBVery high
Q4_K_M *2.32 GB1.8 GB4.1 GB8 GBHigh — the default pick
Q4_12.39 GB1.8 GB4.1 GB8 GBHigh
Q4_K_XL2.37 GB1.8 GB4.1 GB8 GBHigh
Q4_02.21 GB1.8 GB4.0 GB8 GBHigh
Q4_K_S2.21 GB1.8 GB4.0 GB8 GBHigh
IQ4_NL2.2 GB1.8 GB4.0 GB8 GBHigh
IQ4_XS2.11 GB1.8 GB3.9 GB8 GBHigh
Q3_K_XL2 GB1.8 GB3.8 GB8 GBAcceptable — visible loss
Q3_K_M1.95 GB1.8 GB3.7 GB8 GBAcceptable — visible loss
Q3_K_S1.8 GB1.8 GB3.5 GB8 GBAcceptable — visible loss
IQ3_XXS1.59 GB1.8 GB3.3 GB8 GBAcceptable — visible loss
Q2_K_XL1.65 GB1.8 GB3.4 GB8 GBExperimental — not ranked — never recommended
Q2_K1.61 GB1.8 GB3.4 GB8 GBExperimental — not ranked — never recommended
Q2_K_L1.61 GB1.8 GB3.4 GB8 GBExperimental — not ranked — never recommended
IQ2_M1.46 GB1.8 GB3.2 GB8 GBExperimental — not ranked — never recommended
IQ2_XXS1.25 GB1.8 GB3.0 GB8 GBExperimental — not ranked — never recommended
IQ1_M1.15 GB1.8 GB2.9 GB8 GBExperimental — not ranked — never recommended
IQ1_S1.1 GB1.8 GB2.9 GB8 GBExperimental — not ranked — never recommended

* default pick. Weights = real GGUF file sizes from unsloth/gemma-3-4b-it-GGUF (probed 2026-09-02). KV = fp16 estimate; a q8_0 cache roughly halves it. "Comfortable" = weights + KV within 90% of memory.

Best Gemma 3 4B Instruct quant by memory

MemoryRecommended quantTotal (16k ctx)
8 GBQ8_05.6 GB
12 GBBF169.0 GB

Why we don't rank Gemma 3 4B Instruct's 2-bit quants

Quants at 2 bits per weight or below (Q2_K, IQ2, IQ1, TQ1) cut file size by roughly half versus Q4, but the quality collapse is steep and non-linear: perplexity spikes, instruction-following degrades, and hallucinations rise. A model that answers faster but wrong is not a smaller model — it is a worse one. ModelFit lists these builds for completeness but never ranks or recommends them.

Frequently asked questions

What is the best quantization of Gemma 3 4B Instruct?

Q4_K_M is the default pick: 2.32 GB of weights, high — the default pick quality, fitting comfortably in 8 GB of memory (weights + 16k context KV-cache). Go Q6_K or Q8_0 if you have headroom.

How much memory does Gemma 3 4B Instruct need?

At Q4_K_M, Gemma 3 4B Instruct needs 2.32 GB for the weights plus ~1.8 GB of KV-cache at 16k context — about 4.1 GB total, so a 8 GB card or Mac (90% usable budget) runs it comfortably.

Should I use a Q2_K or IQ2 quant of Gemma 3 4B Instruct?

No. Gemma 3 4B Instruct at 2 bits per weight is a visibly worse model — quality collapse at that bitrate is steep, not gradual. If only a 2-bit build fits your memory, run a smaller model at Q4_K_M instead. ModelFit lists these builds but never recommends them.

Cite this page

ModelFit: Gemma 3 4B Instruct quantization comparison (real GGUF file sizes).
https://modelfit.io/quant-compare/gemma3-4b/ (data probed 2026-09-02, CC BY 4.0).