Model benchmarks

google/gemma-4-12b-qat local LLM performance

As of July 2026, google/gemma-4-12b-qat runs at up to 65.9 tok/s for local inference (best of 2 community benchmark runs across 1 GPU).

LM Studio

Model size

12B

Peak speed

65.9 tok/s

Average speed

64.2 tok/s

Min memory

6.7 GB

Max context

6,881 tokens

Best quality

86.6

Benchmark runs

2

GPUs tested

1

Performance by hardware and tool

Every hardware/tool/quantization combination google/gemma-4-12b-qat has been benchmarked on, ranked by peak token generation speed. Last updated July 2026.

HardwareToolQuantBest tok/sAvg tok/sMemoryContextQualityRuns
NVIDIA GeForce RTX 4070 Ti SUPERLM Studio65.9 tok/s64.2 tok/s6.7 GB6,881 tokens71.32

Frequently asked questions

How fast is google/gemma-4-12b-qat for local inference?
Across 2 community benchmark runs, google/gemma-4-12b-qat reaches up to 65.9 tok/s and averages 64.2 tok/s, with the fastest results on NVIDIA GeForce RTX 4070 Ti SUPER.
How much memory does google/gemma-4-12b-qat need?
The leanest observed configuration used about 6.7 GB of memory.
Which tools have been used to run google/gemma-4-12b-qat?
Benchmarks were submitted using LM Studio. Results are community-contributed and updated as new runs arrive.