Gemma 4 family: which size to run
5 sizes, 5.1B–30.7B parameters · Google
Sizes
| Model | Architecture | |||||||
|---|---|---|---|---|---|---|---|---|
| Gemma 4 E2B | 5.1B | Dense | 131,072 | No LiveBench result for this model in release 2026-06-25 | 12* | 6* | 3.1 | 4.3 |
| Gemma 4 E4B | 8B | Dense | 131,072 | No LiveBench result for this model in release 2026-06-25 | 14* | 8* | 5 | 6.8 |
| Gemma 4 12B | 11.95B | Dense | 262,144 | No LiveBench result for this model in release 2026-06-25 | 17 | 11 | 7.1 | 6.7 |
| Gemma 4 26B A4B | 25.2B | MoE · 3.8B active | 262,144 | No LiveBench result for this model in release 2026-06-25 | 27 | 21 | 16.9† | 15.6 |
| Gemma 4 31B | 30.7B | Dense | 262,144 | No LiveBench result for this model in release 2026-06-25 | 32 | 26 | 18.3 | 18.4 |
Each size is its own model, with its own scores and memory needs. Q4 estimates are planning figures, not measurements; file sizes are published. The Mac or PC column is for a 32 GB or larger machine (8 GB reserve); machines under 32 GB reserve 25% of their memory (2 to 8 GB), so they need less. The publisher lists E2B, E4B, 12B, 26B A4B, 31B. * The model's page explains why its estimate differs.
† Unsloth's dynamic UD-Q4_K_M file, shown when the repository has no plain Q4_K_M file (the same rule as the model pages); the cell's title names the file.
How to read this table
Q4 means compressed to about 4 bits per weight. GGUF (llama.cpp, Ollama, LM Studio) and MLX (Apple silicon) are file formats; the sizes shown are the published 4-bit files. A dash means no value is published or collected for that size; the reason is in the cell's accessible text.
Which size fits my machine
The largest size whose Q4 planning estimate fits what each machine offers. A planning estimate, not a measurement and not a speed.
| Machine | Largest Gemma 4 size with enough memory | ||
|---|---|---|---|
| Mac mini M2 · 8 GB unified | 8 | Gemma 4 E4B | 8 |
| Mac mini M4 · 16 GB unified | 16 | Gemma 4 12B | 13 |
| Mac mini M4 · 24 GB unified | 24 | Gemma 4 12B | 15 |
| Mac mini M4 · 32 GB unified | 32 | Gemma 4 31B | 32 |
| Mac mini M5 Pro · 64 GB unified | 56 | Gemma 4 31B | 32 |
| Mac Studio M5 Max · 128 GB unified | 104 | Gemma 4 31B | 32 |
| Mac Studio · M5 Ultra · 36-core CPU / 80-core GPU · 512 GB unified | 392 | Gemma 4 31B | 32 |
| RTX 4060 · 8 GB VRAM | 8 | Gemma 4 E4B | 8 |
| RTX 3060 · 12 GB VRAM | 12 | Gemma 4 12B | 11 |
| RTX 5080 · 16 GB VRAM | 16 | Gemma 4 12B | 11 |
| RTX 4090 · 24 GB VRAM | 24 | Gemma 4 26B A4B | 21 |
| RTX 5090 · 32 GB VRAM | 32 | Gemma 4 31B | 26 |
| DGX H100 · 8 × 80 GB | 640 | Gemma 4 31B | 40 |
A dash means no size of this family fits that machine. The requirement is the size's Q4 estimate with that machine's own reserve: a graphics card reserves 2 GB, as in the sizes table; a Mac or PC reserves 25% of its memory (2 to 8 GB), so below 32 GB its figure is lower than the sizes table's. A server profile reserves 2 GB per graphics card, so its figure is higher.
Smallest machine for each size
The profile with the least memory that still has enough for the size's Q4 estimate, from the hardware catalog.
| Size | Smallest Mac or PC-memory profile | Smallest graphics-card profile |
|---|---|---|
| Gemma 4 E2B | Mac mini 2020 · M1 · 8 GB | RTX 3050 · 6 GB |
| Gemma 4 E4B | Mac mini 2020 · M1 · 8 GB | RTX 3050 · 8 GB |
| Gemma 4 12B | Mac mini 2020 · M1 · 16 GB | RTX 3060 · 12 GB |
| Gemma 4 26B A4B | Mac mini 2024 · M4 · 32 GB | RTX 3090 · 24 GB |
| Gemma 4 31B | Mac mini 2024 · M4 · 32 GB | RTX 5090 · 32 GB |
A dash means no profile of that kind in the catalog has enough memory for the size.
Memory by file precision
The same planning formula at three precisions (0.625, 1.125 and 2 bytes per parameter, plus working room and the system reserve). Estimates, not measurements.
| Size | ||||||
|---|---|---|---|---|---|---|
| Gemma 4 E2B | 12* | 6* | 15* | 9* | 21* | 15* |
| Gemma 4 E4B | 14* | 8* | 19* | 13* | 28* | 22* |
| Gemma 4 12B | 17 | 11 | 25 | 19 | 37 | 31 |
| Gemma 4 26B A4B | 27 | 21 | 43 | 37 | 69 | 63 |
| Gemma 4 31B | 32 | 26 | 50 | 44 | 82 | 76 |
* The model's page explains why its estimate differs.
Family tree
Family, sizes and, for each size, the two 4-bit files the tables use (GGUF Q4_K_M and MLX 4-bit) when one was collected. Other precisions are not drawn here.
The same tree as a list:
- Gemma 4 · Google
- E2B
- GGUF Q4_K_M · 3.1 GB · unsloth/gemma-4-E2B-it-GGUF, read Oct 23, 2026
- MLX 4-bit · 4.3 GB · lmstudio-community/gemma-4-E2B-it-MLX-4bit, read Oct 23, 2026
- E4B
- GGUF Q4_K_M · 5 GB · unsloth/gemma-4-E4B-it-GGUF, read Oct 23, 2026
- MLX 4-bit · 6.8 GB · lmstudio-community/gemma-4-E4B-it-MLX-4bit, read Oct 23, 2026
- 12B
- GGUF Q4_K_M · 7.1 GB · unsloth/gemma-4-12b-it-GGUF, read Oct 23, 2026
- MLX 4-bit · 6.7 GB · lmstudio-community/gemma-4-12B-it-MLX-4bit, read Oct 23, 2026
- 26B A4B · 3.8B active
- GGUF UD-Q4_K_M · 16.9 GB · unsloth/gemma-4-26B-A4B-it-GGUF, read Oct 23, 2026
- MLX 4-bit · 15.6 GB · lmstudio-community/gemma-4-26B-A4B-it-MLX-4bit, read Oct 23, 2026
- 31B
- GGUF Q4_K_M · 18.3 GB · unsloth/gemma-4-31B-it-GGUF, read Oct 23, 2026
- MLX 4-bit · 18.4 GB · lmstudio-community/gemma-4-31B-it-MLX-4bit, read Oct 23, 2026
- E2B
Learn more
See Gemma 4 E2B
The smallest size of the family: its scores, memory estimate and the machines it runs on.
Sources
Where the family and its sizes are named, and where each number comes from.
- google/gemma-4-31B-it model card · “Gemma is a family of open models built by Google DeepMind. Gemma 4 models are multimodal ... The models are available in five distinct sizes: E2B, E4B, 12B, 26B A4B, and 31B.” · read Oct 17, 2026
- Gemma 4 E2B model card
- Gemma 4 E4B model card
- Gemma 4 12B model card
- Gemma 4 26B A4B model card
- Gemma 4 31B model card
- LiveBench 2026-06-25 · the only quality source on this page; a dash means the release does not include the model · snapshot collected Sep 26, 2026
- Published file sizes: Hugging Face repositories, read Oct 23, 2026
- Methodology · how the Q4 planning estimate is computed · Learn