London, September 25 (AI Briefing) -- Koray Kavukcuoglu, newly elevated head of Google DeepMind, confirmed that flagship model Gemini 4 has entered early post-training, covering behaviour refinement and safety testing, and will roll out much earlier than the end of 2026. Speaking at The Information's AI Agenda Live Summit, he noted the milestone is critical as Google works to close the gap with OpenAI and Anthropic, having not shipped a flagship model since Gemini 3 in November 2025.
Internal teams already use Gemini 4 to power the Antigravity coding tool. Separately, Google released Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, rolling out in the Gemini API and AI Studio.
Users can design new voices from a text prompt or choose from more than 2,000 presets, and Flash TTS can recreate a voice from a 30-second sample after a consent check. Google said Flash TTS topped Hume AI's Voice Design Benchmark at 71.4 and that both models placed first and second on Hume's Overall Quality Index, with Figma, HeyGen, Wondercraft and Ollang among the first integrators.
Google Cloud also opened Gemini 3.8 Live with Live Avatar to enterprises, supporting real-time lip synchronisation, automatic switching across 97 languages and asynchronous background tool calls, with SynthID watermarks embedded in generated audio and video.
