Google DeepMind's New Chief Prioritizes Fast Gemini 4 Release Over AGI Debate

Google DeepMind's new chief Koray Kavukcuoglu says Gemini 4 is already in early post-training and he wants it released "much earlier" than the end of 2026, marking a sharp shift in priorities for a lab long defined by its pursuit of artificial general intelligence.

Speaking at The Information's AI Agenda Live Summit in his first media appearance since taking over as head of Google DeepMind, Kavukcuoglu said the team is eager to ship an early post-training version of Gemini 4 rather than wait for a fully polished release. "Our intention is to [...] as soon as possible [...] release an early post-training output because we see the results and we are excited," he said, adding that fast iterations would follow. Google is still finalizing safeguards and safety testing, and engineers are reportedly already using the model internally for the Antigravity coding tool.

A flagship model overdue

Gemini 4 would be Google's first new flagship since Gemini 3 launched late last year. Since then, the company has shipped only the Gemini 3.1 update in February and smaller Flash variants. CEO Sundar Pichai announced Gemini 3.5 Pro at I/O in May for a planned June release, but according to the Financial Times, the project missed three deadlines and never shipped. Kavukcuoglu did not confirm whether Gemini 3.5 Pro has been canceled, saying only that Google took "a little bit of a step back" to "maximize the speed of learning."

In the interim, competitors have pulled ahead. Anthropic released its Mythos/Fable model family in spring 2026 and updated it, alongside the Opus line, this month. OpenAI has been rolling out GPT-6 models since early September. Kavukcuoglu said he remains "a hundred percent confident" in his team, calling it "a certainty" that Google will stay at the frontier — though he offered no benchmark data or timeline detail to support the claim.

Google has also tied its model work to its chip business. The company began selling TPUs directly to customers this year, competing more directly with Nvidia. Kavukcuoglu said new model development informs the design of the "next two or three generations" of chips, according to his comments at the summit.

AGI takes a back seat

The more striking shift is rhetorical. "The conversation of [whether] we achieved AGI or not is not the right conversation," Kavukcuoglu said. "The conversation is more about are we able to build intelligent agents that we can trust?" That stands in contrast to predecessor Demis Hassabis, who wrote in his farewell memo weeks earlier: "I've been working towards AGI my whole life and now, like many of you, I feel it is close at hand."

Hassabis co-founded DeepMind in 2010; Google acquired it in 2014, and it operated with unusual independence until the rise of ChatGPT prompted a 2023 merger with Google Brain into "Google DeepMind." The August 2026 restructuring completed that transition: Kavukcuoglu runs day-to-day operations without a CEO title, Hassabis moved to chairman and Alphabet chief scientist, and Gemini development is consolidating in the Bay Area. A former Google executive told the Guardian: "The era of DeepMind as an independent actor is over."

The Financial Times reports that executives were frustrated by Hassabis prioritizing research openness — such as releasing AlphaFold openly — over commercial pressure. Talent departures have accelerated: Noam Shazeer left for OpenAI in June, AlphaFold Nobel laureate John Jumper joined Anthropic days later, and researchers Jonas Adler and Alexander Pritzel also moved to Anthropic. According to Fortune, DeepMind's new-hire-to-departure ratio fell from roughly 12-to-1 in Q2 2023 to about 2-to-1 in Q3 2026.

What this means

DeepMind's transformation from an AGI-focused research lab into what amounts to Google's Gemini production unit is now largely complete. Kavukcuoglu's comments suggest speed and product delivery have overtaken the open-ended research philosophy that defined the lab's first decade — a shift that mirrors competitive pressure from OpenAI's GPT-6 rollout and Anthropic's rapid model updates. Whether an early, less-polished Gemini 4 can match rivals already in market remains unproven; Google has closed capability gaps before with Gemini 3, but this time it's doing so amid a serious talent exodus and without a confirmed release date. The real signal isn't the timeline — it's that DeepMind's leadership no longer treats AGI as the organizing question.