Gemma 4 QAT models hit LM Studio
Google's Gemma 4 models optimized using Quantization-Aware Training (QAT) are now live in LM Studio. This optimization is available across all sizes of the Gemma 4 model family, allowing users to run these models locally with significantly reduced memory requirements while maintaining high quality and performance.
QAT is a game changer for running large open-weights models locally because it adapts parameters to precision loss during training, outperforming standard post-training quantization.
- –Direct integration into LM Studio lowers the entry barrier for developers and hobbyists trying to run capable models on consumer hardware.
- –Memory footprint reductions enable larger Gemma 4 model variants to run on standard 16GB VRAM systems.
- –This release highlights the shift towards memory-efficient local deployment options as a standard offering alongside base model launches.
DISCOVERED
2h ago
2026-06-05
PUBLISHED
2h ago
2026-06-05
RELEVANCE
AUTHOR
lmstudio