Google DeepMind Chief Says Gemini 4 Could Ship Before Year-End
News Summary
Google DeepMind's new operational chief, Koray Kavukcuoglu, said this week that Gemini 4 is entering its final stretch of development and that the company intends to ship an early post-training version of the model well before the end of the year, rather than waiting for a fully polished release. The comments, made during his first public appearance since taking over day-to-day leadership of the lab, mark the most concrete public signal yet on the timing of Google's next flagship model.
A New Leader, A First Public Statement
Kavukcuoglu made the remarks at The Information's AI Agenda Live Summit in September 2026, Pacific Time, in his debut public appearance as the operational head of Google DeepMind. He described Gemini 4 as having entered the early phase of post-training — the refinement stage that follows a model's base pretraining run, where a model is tuned through techniques like reinforcement learning and human feedback before public release.
"Our intention is to roll out an early post-training version as soon as possible, because we've already seen promising results and are very excited," Kavukcuoglu said, according to reporting from The Information. He indicated Google DeepMind would rather iterate quickly and ship an early checkpoint than hold the model back until it reaches a fully matured state, a strategy aimed at accelerating the cadence at which Google can respond to competitors.
Leadership Reset at Google DeepMind
Kavukcuoglu's comments come roughly a month after a significant leadership reorganization at Google DeepMind. On August 5, 2026, Pacific Time, Google announced that cofounder and longtime CEO Demis Hassabis would step back from day-to-day management, moving into the newly defined role of Chair of Google DeepMind and Chief Scientist of Alphabet, while continuing to lead the drug-discovery venture Isomorphic Labs. Kavukcuoglu, who had been serving as Chief AI Architect reporting directly to Google CEO Sundar Pichai since mid-2025, was elevated to Senior Vice President of Google DeepMind, taking direct responsibility for Gemini model development, frontier AI research, and the teams behind the Gemini app and developer platform.
Hassabis has said the shift allows him to focus on longer-horizon questions around AI safety, governance, and scientific applications as the field moves closer to advanced general-purpose AI systems, while day-to-day execution on Gemini shifts to Kavukcuoglu, who has been with DeepMind since its early years and has long overseen reinforcement learning research and Gemini's technical development. The same reorganization saw Alphabet and Google chief scientist Jeff Dean depart the company after 27 years to start a new AI-focused venture.
Why the Timeline Matters
Google has faced sustained pressure to keep pace with rival AI labs, particularly OpenAI and Anthropic, both of which have shipped frequent model updates over the past year. Industry observers have framed Kavukcuoglu's accelerated timeline as an attempt to close that competitive gap: shipping an early, iterating version of Gemini 4 before the end of 2026 would reinforce the narrative that Google is catching up to or matching its rivals' pace, while a delay into 2027 would instead suggest Google continues to trail the field on release cadence.
Alphabet's stock saw volatility around the commentary, a reminder of how closely investors are tracking Google's AI execution alongside its cloud and search businesses. Kavukcuoglu did not commit to an exact release date, emphasizing instead that the current post-training checkpoint is showing results the team is confident enough in to bring to market ahead of a fully finished version.
What Comes Next
No specific launch date has been confirmed for Gemini 4. Google has historically used developer events and its own blog to formally announce new Gemini generations, and the company is expected to provide further details as the post-training process progresses. For now, Kavukcuoglu's remarks give the clearest public indication that Google DeepMind, under its new leadership structure, is prioritizing speed of iteration over waiting for a single, fully polished model release — a notable shift in strategy as the broader AI industry continues to compress the time between major model releases.