Google announces Gemini 4's post-training phase and potential early release

At The Information's AI Agenda Live event on September 23, Demis Hassabis, head of Google DeepMind, confirmed that the next-generation flagship model Gemini 4 has entered the 'early post-training' phase. Hassabis expressed optimism about the model's release, stating it could happen 'well before the end of the year.' He emphasized the team's excitement over early results from the post-training stage and highlighted ongoing rapid iteration. Gemini 4 marks Google's first next-generation flagship since the Gemini 3 series, which was launched at the end of last year. In February, Google released an upgraded version of Gemini 3.1 and introduced smaller Flash models. Hassabis explained that after the release of Gemini 3 and 3.1, Google shifted focus to Flash models, prioritizing 'accelerating learning.' He did not confirm whether the previously announced Gemini 3.5 Pro was canceled but stressed the current emphasis on launching Gemini 4. When asked about competition, Hassabis expressed confidence in his team, stating, 'we will always be at the forefront of technology.' Gemini 4 is currently undergoing safety testing and has been used internally to run the Antigravity AI programming tool. Hassabis avoided commenting on whether the industry has achieved AGI, instead emphasizing the need to build 'trustworthy agents.'