Google Releases Gemma 4 12B: Encoder-Free Multimodal Model With 256K Context Tuned for Mobile

Mervin Praison·low signal

Google's Gemma 4 12B arrived under Apache 2.0 as an encoder-free multimodal model with a 256K context window and quantization-aware training (QAT) optimized for on-device/mobile deployment. The encoder-free design simplifies the multimodal stack while the QAT path targets running locally on constrained hardware. For builders, it's a permissively licensed, genuinely deployable multimodal base for edge and offline apps rather than another cloud-only release.