Google/Mixture of Experts

GoogleGemma 4 26B A4B

Gemma 4 26B A4B — MoE with 128 experts, 4B active. Matches dense 31B at fraction of compute. 256K context.

chatcodingreasoningmultilingualvision
26B
Parameters (4B active)
256K
Context length
7
Benchmarks
10
Quantizations
300K
HF downloads
Architecture
MoE
Released
2026-04-02
Layers
30
KV Heads
8
Head Dim
256
Family
gemma

Quantization Options

QuantBitsVRAMQuality
Q3_K_M413.5 GBlow
Q3_K_L4.314.5 GBmoderate
IQ4_XS4.4615.0 GBmoderate
Q4_K_S4.6715.7 GBmoderate
Q4_K_M4.8916.4 GBgood
Q5_K_S5.5718.6 GBgood
Q5_K_M5.719.0 GBgood
Q6_K6.5621.8 GBexcellent
Q8_08.528.1 GBlossless
FP161652.5 GBlossless

Select your GPU above to see speed estimates and compatibility for each quantization.

Benchmarks (7)

Arena Elo1441
AIME88.3
MMLU-PRO82.6
GPQA Diamond82.3
LiveCodeBench77.1
BBH64.8
HLE8.7

Run this model

Easiest way to get starteddocs →
curl -fsSL https://ollama.com/install.sh | sh
$ollama run gemma4:26b:q4_k_m

Downloads and runs automatically. Add --verbose for speed stats.

Setup guide

GPUs that can run this model

At Q4_K_M quantization. Sorted by minimum VRAM.

Apple M3 Pro (18GB)
18 GB VRAM • 150 GB/s
APPLE
$1599
AMD RX 7900 XT
20 GB VRAM • 800 GB/s
AMD
$849
NVIDIA RTX 4000 Ada 20GB
20 GB VRAM • 432 GB/s
NVIDIA
$1250
NVIDIA A10M
20 GB VRAM • 500 GB/s
NVIDIA
NVIDIA GeForce RTX 3080 Ti 20 GB
20 GB VRAM • 760 GB/s
NVIDIA
$1199
AMD Radeon RX 7900 XT
20 GB VRAM • 800 GB/s
AMD
$899
NVIDIA RTX 4000 Ada Generation
20 GB VRAM • 360 GB/s
NVIDIA
$1250
NVIDIA RTX 4000 SFF Ada Generation
20 GB VRAM • 280 GB/s
NVIDIA
$1250
NVIDIA RTX A4500
20 GB VRAM • 640 GB/s
NVIDIA
$2000
NVIDIA RTX 4090
24 GB VRAM • 1008 GB/s
NVIDIA
$1599
NVIDIA RTX 3090 Ti
24 GB VRAM • 1008 GB/s
NVIDIA
$999
NVIDIA RTX 3090
24 GB VRAM • 936 GB/s
NVIDIA
$850
AMD RX 7900 XTX
24 GB VRAM • 960 GB/s
AMD
$999
Apple M4 Pro (24GB)
24 GB VRAM • 273 GB/s
APPLE
$1399
NVIDIA L4 24GB
24 GB VRAM • 300 GB/s
NVIDIA
$2500
NVIDIA A10 24GB
24 GB VRAM • 600 GB/s
NVIDIA
$3500
Apple M2 (24GB)
24 GB VRAM • 100 GB/s
APPLE
$999
Apple M3 (24GB)
24 GB VRAM • 100 GB/s
APPLE
$999
Apple M4 (24GB)
24 GB VRAM • 120 GB/s
APPLE
$699
NVIDIA Tesla M40 24 GB
24 GB VRAM • 288 GB/s
NVIDIA
NVIDIA Tesla P10
24 GB VRAM • 694 GB/s
NVIDIA
NVIDIA Tesla P40
24 GB VRAM • 347 GB/s
NVIDIA
NVIDIA Quadro RTX 6000
24 GB VRAM • 672 GB/s
NVIDIA
$4000
NVIDIA Quadro RTX 6000 Passive
24 GB VRAM • 624 GB/s
NVIDIA
$4000
NVIDIA GeForce RTX 3090
24 GB VRAM • 936 GB/s
NVIDIA
$1499
NVIDIA A10 PCIe
24 GB VRAM • 600 GB/s
NVIDIA
NVIDIA A10G
24 GB VRAM • 600 GB/s
NVIDIA
NVIDIA RTX A5000
24 GB VRAM • 768 GB/s
NVIDIA
$2500
NVIDIA GeForce RTX 3090 Ti
24 GB VRAM • 1010 GB/s
NVIDIA
$1999
NVIDIA GeForce RTX 4090
24 GB VRAM • 1010 GB/s
NVIDIA
$1599

Find the best GPU for Gemma 4 26B A4B

Build Hardware for Gemma 4 26B A4B

Gemma 4 26B A4B26B Parameter Mixture of Experts LLM

Model Specifications

Parameters
26B (4B active)
Architecture
Mixture of Experts
Context Length
256K tokens
Capabilities
chat, coding, reasoning, multilingual, vision
Release Date
2026-04-02
Provider
Google
Family
gemma

VRAM Requirements

QuantizationBPWVRAMQuality
Q3_K_M413.5 GB88%
Q3_K_L4.314.5 GB90%
IQ4_XS4.4615.0 GB92%
Q4_K_S4.6715.7 GB93%
Q4_K_M4.8916.4 GB94%
Q5_K_S5.5718.6 GB96%
Q5_K_M5.719.0 GB96%
Q6_K6.5621.8 GB97%
Q8_08.528.1 GB100%
FP161652.5 GB100%

Benchmark Scores

MMLU-PRO82.6
BBH64.8
Arena Elo1441.0
GPQA Diamond82.3
HLE8.7
LiveCodeBench77.1
AIME88.3

How to Run Gemma 4 26B A4B

Run Gemma 4 26B A4B locally with Ollama (needs 16.4 GB VRAM at Q4_K_M):

ollama run gemma4:26b

Compatible GPUs (30)

GPUs that can run Gemma 4 26B A4B at Q4_K_M quantization: