canirunaimodel
Google · 27B (4B active) · Mixture of Experts
Gemma 4 MoE base model (official)
Use Cases
Mixture of Experts
| Quant | Bits | VRAM | Quality | Status |
|---|---|---|---|---|
| Q2_K | 2 | 9.1 GB | low | — |
| Q3_K_M | 3 | 12.6 GB | moderate | — |
| Q4_K_M | 4 | 14.3 GB | good | — |
| Q5_K_M | 5 | 17.8 GB | good | — |
| Q6_K | 6 | 21.2 GB | excellent | — |
| Q8_0 | 8 | 28.2 GB | excellent | — |
| F16 | 16 | 55.8 GB | lossless | — |
About this model
Mixture of Experts model for vision, reasoning workloads.
Context
256K
Q4 VRAM
~14.3 GB
Gemma 4 26B-A4B on your hardware
Gemma 4 MoE base model (official). This page turns the Hugging Face model card into practical local-run numbers, so you can compare quantized VRAM, system RAM, and expected fit before downloading a large checkpoint.
27B total parameters with 4B active. 4 of 26 experts are active per token.
Start with Q4_K_M: about 13.2 GB on disk and 14.3 GB VRAM before extra context and runtime overhead.
vision, reasoning
Listed as Gemma from huggingface.co/google/gemma-4-26B-A4B.
Want the real verdict? Pick your GPU or edit the specs on this page and compare the quant table below.
Open HF repo