Skip to content

Gemma 4 26B A4B

MoEVision

GoogleGoogle · 25.8B (3.8B active) · Mixture of experts

Gemma 4 26B A4B is a 25.8B model from Google with a 256K-token context. At Q4_K_M with an 8K context it needs about 16.5 GB; 22 GB leaves room for longer chats.

Hugging Face GGUF

Mixture of experts

Parameters: 25.8B · Active: 3.8B

All 25.8B parameters load into memory, but only 3.8B work on each token, so it runs at the speed of a much smaller model.

Quantization options

QuantMemoryOn your hardware
Q2_K—…
Q3_K_M—…
Q4_K_M—…
Q5_K_M—…
Q6_K—…
Q8_0—…
F16—…

Memory by context length

At Q4_K_M. The context cache grows with every token the model keeps in mind.

ContextContext cacheTotal
4K0.3 GB16.4 GB
8K0.4 GB16.5 GB
32K0.8 GB17 GB
128K2.7 GB18.9 GB
256K5.2 GB21.4 GB

Share a measured speed

Run one of these and paste the whole output below (or just the tokens per second):

llama-bench -m model.gguf
ollama run model --verbose

Published anonymously and kept. We store no account or address, only a daily-changing hash for a limit of ten submissions a day.

Next steps

Smaller alternatives

Models that need less memory than Gemma 4 26B A4B and score about as well or better.

Can I run Gemma 4 26B A4B locally?

Can I run Gemma 4 26B A4B locally?

Yes, if your GPU or Mac has about 16.5 GB free for it at Q4_K_M. Open this page on that computer to see the grade for your exact hardware, then start it with llama.cpp, Ollama or LM Studio.

How much memory does Gemma 4 26B A4B need?

About 16.5 GB at Q4_K_M with an 8K context and 25.7 GB at Q8_0. Lower quantizations fit smaller cards with some loss in quality; longer contexts add to the total.