1/ In Google DeepMind’s news list, as of September 2026, the items Gemini 3.8 Flash, Gemini 3.8 Live, and Gemini 3.8 Live Extended Thinking newly appear, and improvements to Private AI Compute including secure server-side memory are also confirmed.
2/ However, since that page shows titles and categories before the detailed body of each announcement, at this point only the fact that a related official announcement was made in September can be confirmed.
3/ Such list updates show that DeepMind is advancing multimodal and real-time interactive models together with privacy features.
4/ The simultaneous strengthening of model performance and memory/security functions suggests that data protection requirements are becoming more important in actual product deployment and in corporate and research use. [1]
Implications
5/ For materials and science researchers, improvements in real-time interactive models can increase convenience for tasks such as experimental note assistance, meeting note organization, experimental condition Q&A, and research tool integration.
6/ In particular, if secure execution and server-side memory improvements proceed together, applicability in environments handling sensitive research data may also increase.
7/ However, since only list-level information has been confirmed, actual performance improvements and data handling methods require additional verification through individual announcements.
8/ Research teams should also check privacy, retention policies, access control, and reproducibility impact before adoption. [1]