Gemma 4 26B-A4B quants compared

27 GGUF builds by real file size, probed from bartowski/google_gemma-4-26b-a4b-it-GGUF on Hugging Face (2026-09-02). 26B params, 4B active.

Download Gemma 4 26B-A4B Q4_K_M (15.87 GB) — it fits 24 GB of memory with 16k context. With Ollama: ollama run gemma4:26b

Every Gemma 4 26B-A4B quant by real file size

QuantWeights+ KV (16k)TotalFits comfortably inQuality
BF1647.04 GB4.0 GB51.0 GB64 GBFull precision (lossless)
Q8_025.45 GB4.0 GB29.4 GB48 GBNear-lossless
Q6_K_L21.46 GB4.0 GB25.5 GB32 GBExcellent
Q6_K21.29 GB4.0 GB25.3 GB32 GBExcellent
Q5_K_L18.16 GB4.0 GB22.2 GB32 GBVery high
Q5_K_M17.99 GB4.0 GB22.0 GB32 GBVery high
Q5_K_S16.88 GB4.0 GB20.9 GB24 GBVery high
Q4_K_M *15.87 GB4.0 GB19.9 GB24 GBHigh — the default pick
Q4_K_L16.03 GB4.0 GB20.0 GB24 GBHigh
Q4_115.04 GB4.0 GB19.0 GB24 GBHigh
Q4_K_S14.76 GB4.0 GB18.8 GB24 GBHigh
Q4_014.04 GB4.0 GB18.0 GB24 GBHigh
IQ4_NL13.69 GB4.0 GB17.7 GB24 GBHigh
IQ4_XS13.23 GB4.0 GB17.2 GB24 GBHigh
Q3_K_XL12.46 GB4.0 GB16.5 GB24 GBAcceptable — visible loss
IQ3_M12.37 GB4.0 GB16.4 GB24 GBAcceptable — visible loss
Q3_K_L12.29 GB4.0 GB16.3 GB24 GBAcceptable — visible loss
Q3_K_M12.13 GB4.0 GB16.1 GB24 GBAcceptable — visible loss
Q3_K_S11.69 GB4.0 GB15.7 GB24 GBAcceptable — visible loss
IQ3_XS11.58 GB4.0 GB15.6 GB24 GBAcceptable — visible loss
IQ3_XXS11.33 GB4.0 GB15.3 GB24 GBAcceptable — visible loss
Q2_K_L10.37 GB4.0 GB14.4 GB16 GBExperimental — not ranked — never recommended
Q2_K10.2 GB4.0 GB14.2 GB16 GBExperimental — not ranked — never recommended
IQ2_M9.97 GB4.0 GB14.0 GB16 GBExperimental — not ranked — never recommended
IQ2_S9.53 GB4.0 GB13.5 GB16 GBExperimental — not ranked — never recommended
IQ2_XS9.43 GB4.0 GB13.4 GB16 GBExperimental — not ranked — never recommended
IQ2_XXS8.99 GB4.0 GB13.0 GB16 GBExperimental — not ranked — never recommended

* default pick. Weights = real GGUF file sizes from bartowski/google_gemma-4-26b-a4b-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 4 26B-A4B quant by memory

MemoryRecommended quantTotal (16k ctx)
24 GBQ5_K_S20.9 GB
32 GBQ6_K25.3 GB
48 GBQ8_029.4 GB
64 GBBF1651.0 GB

Why we don't rank Gemma 4 26B-A4B'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 4 26B-A4B?

Q4_K_M is the default pick: 15.87 GB of weights, high — the default pick quality, fitting comfortably in 24 GB of memory (weights + 16k context KV-cache). Go Q6_K or Q8_0 if you have headroom.

How much memory does Gemma 4 26B-A4B need?

At Q4_K_M, Gemma 4 26B-A4B needs 15.87 GB for the weights plus ~4.0 GB of KV-cache at 16k context — about 19.9 GB total, so a 24 GB card or Mac (90% usable budget) runs it comfortably.

Should I use a Q2_K or IQ2 quant of Gemma 4 26B-A4B?

No. Gemma 4 26B-A4B 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 4 26B-A4B quantization comparison (real GGUF file sizes).
https://modelfit.io/quant-compare/gemma4-26b-a4b/ (data probed 2026-09-02, CC BY 4.0).