Community performance summary

Gemma4 26B-A4B

4 GPU configurations measured with this model. Compare hardware differences, quantizations and every recorded result.

4 signed runs · 0 verified · 2 testers
Median prompt517.82tok/s · 46.59 to 1758.66
Median generation22.7tok/s · 5.51 to 50.89
Configurations2Q4_K_M · Q8_K
Last observedSep 22GCN / Vega · RDNA 1 · RDNA 2 · RDNA 2 + GCN / Vega

What the results show

One clear view.
Every recorded run.

Measurements from comparable hardware and models are presented together. Medians show typical performance, while every signed run remains available below for inspection and sharing.

Hardware tested4 groups
AMD Radeon RX 5801 run→AMD Radeon Pro 5600M1 run→AMD Radeon RX 6900 XT ×21 run→AMD Radeon RX 6900 XT + AMD Radeon Vega1 run→
macOS

macOS 15.7.3 Sequoia · macOS 15.7.5 Sequoia · macOS 15.7.9 Sequoia

Recorded evidence

Inspect every run.

Open any result to review its exact hardware, model configuration, measurements and submission details.

1 to 8 of 4 configurations

Verified benchmark Community submission
Model / quantHardwareArchitecturePromptGenerationSource
Gemma4 26B-A4BQ4_K_M · MoE · Signed ToshLLM app · amd-gpu · Metal · comparison
RDNA 2
1758.66tok/s
50.89tok/s
ToshLLM communitySep 6, 2026
Gemma4 26B-A4BQ4_K_M · MoE · Signed ToshLLM app · amd-gpu · Metal · comparison
RDNA 2 + GCN / Vega
883.21tok/s
30.44tok/s
ToshLLM communityAug 21, 2026
Gemma4 26B-A4BQ4_K_M · MoE · Signed ToshLLM app · ncmoe 20 · amd-gpu · Metal · comparison
RDNA 1
152.43tok/s
14.95tok/s
ToshLLM communitySep 8, 2026
Gemma4 26B-A4BQ8_K · MoE · Signed ToshLLM app · ncmoe 40 · amd-gpu · Metal · comparison
GCN / Vega
46.59tok/s
5.51tok/s
cloudphreniaSep 22, 2026