[Google Open-Sources Gemma 4 with New 12B Version, Targeting 16GB RAM Laptops] According to monitoring by Beating, Google has released Gemma 4 12B, the first new model specification added to the Gemma 4 family since its launch in April this year. Unlike the previously released ultra-lightweight edge models E2B and E4B, as well as the high-performance inference models 31B and 26B-A4B, the new version is designed for local deployment on consumer-grade computers, specifically targeting multimodal Agents running on laptops with 16GB of RAM. Gemma 4 12B adopts a unified encoder-free multimodal architecture, supporting both text and image inputs. Google has also simultaneously upgraded its LiteRT-LM local inference tool, adding a new OpenAI API-compatible service mode. Developers can directly connect tools like Continue, Aider, and Open WebUI to the locally running Gemma 4 12B without relying on cloud-based models. The entire Gemma 4 model family was officially launched in April, so this is not a next-generation model update but rather an expansion of the product line. Previously, Gemma 4 covered a parameter range of approximately 8B to 31B, with the 12B version filling the gap between performance and hardware requirements. Instead of pursuing larger parameter scales, Google is focusing on how to enable multimodal models with Agent capabilities to run stably on standard laptops. [Original Link]