community pool / model answer

gemma-4-26B-A4B-it-GGUF

chat · unsloth/gemma-4-26B-A4B-it-GGUF

89.9 tok/s · model median · 66 runs in the model record.

Reference rig	Quantization	Median result	Basis	TTFT (ms)	Peak VRAM	Max context	Engines
Arc Pro B70 32GB	8-bit	97.6 tok/s	measured	602	- GB	32,768	llama.cpp
GB10 Grace Blackwell 128GB	4-bit	50.7 tok/s	reported	-	- GB	262,144	llama.cpp
GB10 Grace Blackwell 128GB	8-bit	38.8 tok/s	reported	-	- GB	262,144	llama.cpp
M5 Pro 24GB	4-bit	55 tok/s	reported	446	- GB	4,224	llama.cpp
Minisforum UM790 Pro 64GB	4-bit	19.5 tok/s	reported	-	- GB	2,048	llama.cpp
Radeon RX 7900 XTX 24GB	4-bit	94.3 tok/s	reported	-	15.3 GB	640	llama.cpp
RTX 3060 12GB ×2	4-bit	68.2 tok/s	measured	2,495	18.6 GB	131,072	llama.cpp
RTX 5060 Ti 16GB	3-bit	103.4 tok/s	reported	148	- GB	2,048	llama.cpp
RTX 5090 32GB	3-bit	255.7 tok/s	reported	41	- GB	2,048	llama.cpp
Ryzen AI Max 395 64GB	4-bit	49.2 tok/s	reported	1,004	17.8 GB	8,192	llama.cpp

Context tested

The table above is the tested context for each exact rig and quantization cell. Empty fields remain empty rather than being filled with an estimate.