Gemma 4 Turbo 12b is the balanced performance edition of the Gemma 4 Turbo family — Google's encoder-free 12B parameter model, optimized for CPU via turboquant inference tuning with KV cache quantization (Stage 1 TurboQuant). Delivers the best balance of speed, capability, and memory usage in the family, with native multimodal support for text, images, and audio. No GPU required.