What hardware runs Gemma 4 26B A4B?

Gemma 4 26B A4B is a 25.23B-parameter mixture-of-experts model with 3.8B active parameters (apache-2.0 license). Google's fast MoE with native audio in. Nearly all of its weight sits in routed experts, so expert offload runs it comfortably on 12 GB cards.

All figures assume an f16 KV cache, a 0.6 GB display reserve on the GPU, and 64 GB of DDR5 system RAM for the offload tiers. Tune these in the calculator.

Minimum VRAM by quant and context

Full-GPU figures: weights + f16 KV cache + overhead. * below our recommended floor of Q4_K_M.

GPU compatibility

Fits on GPUExpert offloadPartial offloadCPU only

Quant guidance

Our floor for Gemma 4 26B A4B is Q4_K_M — below that, quality degrades faster than the VRAM savings are worth. Prefer the highest quant that still lands "Fits on GPU" or “Expert offload” at your context length in the table above.

Recommended run command

Q4_K_M at 32K on a NVIDIA GeForce RTX 5070-class GPU (Expert offload):

llama-server -m google_gemma-4-26B-A4B-it-Q4_K_M.gguf -c 32768 --flash-attn -ngl 99 --n-cpu-moe 30