DeepMind was created with the goal of pursuing AGI, but its new leader is focused on simply getting Gemini 4 shipped

Google's new AI chief says an early preview of Gemini 4 could arrive "as soon as possible," with fast iterations to follow. The statement from Oriol Kavukcuoglu, who took over day-to-day leadership of Google Deepmind in August, comes as the company tries to close a widening gap with Anthropic and OpenAI — and as the lab he now runs quietly drops the AGI mission that once defined it.

A Missing Flagship and Three Missed Deadlines

Gemini 4 would be Google's first new flagship model since Gemini 3 launched late last year. In between, the company shipped the Gemini 3.1 update in February and several smaller Flash models. But the bigger stumble came after CEO Sundar Pichai announced Gemini 3.5 Pro at I/O in May for a June release — a model that never arrived. According to the Financial Times, the project missed three deadlines.

Kavukcuoglu admitted Google took "a little bit of a step back," saying the priority was to "maximize the speed of learning at the time." He did not confirm whether Gemini 3.5 Pro has been formally canceled. Engineers, meanwhile, are already using the new model internally for the coding tool Antigravity, while safeguards and safety testing continue ahead of any public release.

Rivals Pull Ahead as Talent Walks Out

The delays have cost Google ground. Anthropic released its Mythos/Fable family in the spring and updated it, along with the Opus line, this month. OpenAI has been rolling out GPT-6 models since early September — and all of them, by most accounts, outperform Google's current top models.

The competitive pressure coincides with a talent drain. Noam Shazeer left for OpenAI in June; two days later, AlphaFold Nobel laureate John Jumper announced a move to Anthropic. Jeff Dean co-founded the startup Discovery Loop with Oriol Vinyals, Quoc Le, and Sanjay Ghemawat, and Gemini researchers Jonas Adler and Alexander Pritzel also joined Anthropic. Per Fortune, Deepmind's ratio of new hires to departures fell from about 12 to 1 in Q2 2023 to roughly 2 to 1 in Q3 2026.

Kavukcuoglu remains unbothered: he is "a hundred percent confident" in his "insanely strong team" and says it is "a certainty that we are always gonna be at the frontier." Google has caught up before — Gemini 3 was such a leap that OpenAI declared a "Code Red" and only pulled even again with GPT-5.2.

AGI Rhetoric Gives Way to Trustworthy Agents

The leadership change itself marks a turning point. Kavukcuoglu, previously CTO, took over in August after Demis Hassabis stepped down as Deepmind CEO; Hassabis now chairs the unit and serves as Alphabet's chief scientist. "The conversation of [whether] we achieved AGI or not is not the right conversation," Kavukcuoglu said — a striking contrast to Hassabis's farewell memo, in which he wrote that AGI "is close at hand." The new chief frames the goal instead as building "intelligent agents that we can trust."

All Gemini development is moving to the Bay Area. "The era of DeepMind as an independent actor is over," a former Google executive told the Guardian. According to the Financial Times, executives were frustrated that Hassabis weighed commercial needs too lightly — as when he released AlphaFold openly. The newsletter Pathfounders, citing industry sources, reports Hassabis wanted to leave alongside Jeff Dean, but management feared a stock price crash. Google hasn't commented.

The timing matters: the AGI debate is louder than it has been in years. On September 12, Anthropic CEO Dario Amodei published "We Must Pace the Frontier," urging frontier labs to slow capability advances. OpenAI followed suit, Elon Musk agreed, and Hassabis called the direction right but the details unfinished. Back in July, more than 1,200 frontier-lab employees signed a letter of the same name.

Kavukcuoglu isn't dismissing safety — trustworthy agents are a safety concern in their own right. His model work also feeds Google's chip business: the company began selling TPUs directly to customers this year, competing more directly with Nvidia, and new models show hardware engineers how to design the "next two or three generations" of chips.

Where Things Stand

Google Deepmind now operates as Google's Gemini factory rather than an independent research lab with an AGI mission — a shift that looks complete. The immediate question is whether Gemini 4's early release, once safety testing wraps, can repeat Gemini 3's frontier-resetting impact. Watch for the post-training preview and how it measures up against GPT-6 and Anthropic's latest Opus models.

Meta description: Google's new Deepmind chief Oriol Kavukcuoglu says an early Gemini 4 release is coming as the lab completes its shift from AGI research to product unit.

Tags: Google Deepmind, Gemini 4, Oriol Kavukcuoglu, Demis Hassabis, AI competition

Featured image: Abstract dark-blue and gold gradient with flowing circuit-like light trails, no people or logos.