PANews reported on April 3rd that Google has officially launched the Gemma 4 open-source large model series, including four models: E2B, E4B, 26B MoE, and 31B. This series is optimized for advanced inference and agent workflows, significantly improving intelligence per unit parameter. The 31B model ranks third globally among open-source models on the Arena AI text leaderboard, outperforming models with 20 times more parameters of the same size. E2B and E4B support offline operation on mobile devices, compatible with smartphones, Jetson, and other devices; the entire series supports 140+ languages, 128K–256K long context, and multimodal input. Developers can integrate applications through the AICore preview version.