gemma4-26b-a4b
Architecture mixture of experts30 LayersLicense apache-2.0Maximum context 256kVocabulary 262,144
At Q4_K_M it is a 15.78 GiB file, and it fits on 12 of the 18 machines we measure (the one with the least memory: Mac with M4 Pro and 24 GB). Here the file is what fills the memory: even at its maximum context of 256k tokens, the context adds 5.20 GiB — and the same content in German takes ×1.33 the tokens of English.
Model size at Q4_K_M
Measured by Local AI Scope
Context memory at 8k tokens
Measured by Local AI Scope
What German costs versus English
Measured by Local AI Scope
Sizes of the real published files
These are the sizes of the files actually published by the author, added up when the model ships split into parts. Not an estimate from the parameter count.
| Compression | Model size |
|---|---|
Q8_0 |
25.02 GiB |
Q6_K |
21.58 GiB |
Q5_K_M |
19.70 GiB |
Q5_K_S |
17.56 GiB |
Q4_K_M |
15.78 GiB |
Q4_K_S |
15.36 GiB |
IQ4_XS |
12.66 GiB |
Q3_K_M |
11.85 GiB |
This model also reads images. For that it needs a second file, the vision projector (mmproj-F16.gguf, 1.11 GiB), published next to the model and not included in the figures on this page: they count text only. If you are going to send it images, add it to the model size.
Context memory
Computed from the attention pattern this model declares layer by layer, not from the generic formula. That is why the figure is often far smaller than other calculators tell you. Its full-attention layers are built differently from its sliding-window ones: 2 key-value heads of 512 per layer instead of 8 key-value heads of 256. We count each type with its own size, and with key and value stored separately, which is what llama.cpp and MLX keep in memory.
- 8k tokens → 0.35 GiB
- 32k tokens → 0.82 GiB
- 128k tokens → 2.70 GiB
What it costs in each language
Measured by running this model’s own tokenizer over the same content in four languages. Tokens are what you pay for in context, in memory and in your cloud bill.
| Kind of text | EN | ES | FR | DE |
|---|---|---|---|---|
| Code | — | ×1.16 | ×1.25 | ×1.39 |
| Contracts | — | ×1.15 | ×1.22 | ×1.36 |
| Business email | — | ×1.11 | ×1.21 | ×1.21 |
| Support tickets | — | ×1.20 | ×1.40 | ×1.32 |
English is the baseline: every figure is how many times more tokens the same content costs in that language. Average across the whole corpus: DE ×1.33.
Where it fits
Fits means the model, its context memory and the working headroom all fit at the stated context, leaving the system its share. 12 of the 18 machine profiles tested (Q4_K_M, 8k tokens).
If you are willing to compress it more, it does run where the table says no: at Q3_K_M (11.85 GiB), up to 8k tokens of context, on Mac with M4 and 16 GB unified memory · Office laptop, CPU only, 16 GB · Mac mini with M6 and 16 GB. Below Q4_K_M the model loses a little quality, and at Q3_K_M you can tell; that is why the table above sticks to Q4_K_M.