Google DeepMind's new leadership has broken its silence. Koray Kavukcuoglu, who took over the lab after Demis Hassabis stepped down in August, made his first media appearance as head of Google DeepMind at The Information's AI Agenda Live Summit on Wednesday and confirmed that the company's next flagship model, Gemini 4, is in the early phase of post-training, the stage where a base model is refined to behave reliably.

Speaking at the summit, Kavukcuoglu said he hopes to release Gemini 4 "much earlier" than the end of the year. "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, according to a report by The Decoder. Fast iterations would follow, with Google still working on safeguards and safety testing. For more context on this story, see our ongoing artificial intelligence updates.

Engineers Are Already Using Gemini 4 Internally

The model is not just a lab experiment. Kavukcuoglu revealed that Google engineers are already using Gemini 4 internally for the company's coding tool, Antigravity, an early signal that the model's agentic coding capabilities are far enough along to be useful in production workflows inside Google itself.

Gemini 4 will be Google's first new flagship model since Gemini 3 launched late last year. Since then, the company has shipped the Gemini 3.1 update in February and a series of smaller, cheaper Flash models.

The Gemini 3.5 Pro Debacle Hangs Over the Launch

The run-up to Gemini 4 carries the baggage of its vanished predecessor. CEO Sundar Pichai announced Gemini 3.5 Pro at Google I/O in May for a June release, but the model never arrived. According to the Financial Times, the project missed three deadlines.

Kavukcuoglu did not say whether Gemini 3.5 Pro has been officially canceled, but acknowledged that Google took "a little bit of a step back." "For us, what was important is to maximize the speed of learning at the time," he explained.

The pause has cost Google ground. Rivals have moved ahead in the meantime: Anthropic released its Mythos/Fable family in the spring and updated it, along with the Opus line, this month, while OpenAI has been rolling out GPT-6 models since early September. All of them are far more capable than Google's current top models, according to The Decoder's assessment.

Kavukcuoglu says he is not worried. He described himself as "a hundred percent confident" and expressed trust in his "insanely strong team." "In my mind, it's a certainty that we are always gonna be at the frontier," he said.

Google has caught up before. Gemini 3 was such a leap that OpenAI reportedly declared a "Code Red" and only pulled even again with GPT-5.2. Whether Gemini 4 can repeat that feat remains an open question.

AGI Is 'Not the Right Conversation'

Just as revealing as the timeline is what the new chief said about AGI, the goal that defined DeepMind for over a decade. "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 is a striking pivot for a lab whose co-founder sounded very different only weeks ago. "I've been working towards AGI my whole life and now, like many of you, I feel it is close at hand," Hassabis wrote in his farewell memo when he stepped down as CEO. Hassabis now chairs the unit and serves as Alphabet's chief scientist, while Kavukcuoglu, previously the company's CTO, runs day-to-day operations.

The restructuring runs deep. According to The Decoder, all Gemini development is moving to the Bay Area, away from the London headquarters that gave DeepMind its identity. "The era of DeepMind as an independent actor is over," a former Google executive told The Guardian. The Financial Times reported that executives were frustrated that Hassabis gave commercial needs less weight, citing his decision to release AlphaFold openly as an example.

The Chip Connection

The model work feeds Google's growing hardware ambitions. The company started selling its TPUs directly to customers this year, putting it in closer competition with Nvidia. According to Kavukcuoglu, new models show hardware engineers how to design the "next two or three generations" of chips, a feedback loop that ties Google's model roadmap directly to its data center economics.

For now, the message from DeepMind's new leadership is simple: less talk about AGI milestones, more shipping. Whether Gemini 4 arrives "much earlier" than year-end, as promised, will be the first real test of whether the restructured lab can move at the pace its rivals have set.

From London Lab to Gemini Factory

The reshaping of DeepMind is the culmination of a longer arc. Hassabis co-founded DeepMind in 2010, and Google acquired the London lab four years later. For years the team enjoyed special status inside the company, chasing fundamental research with unusual freedom. ChatGPT changed that. During Google's first "Code Red," the company merged Google Brain and DeepMind into "Google DeepMind," which is also where the Gemini models got their name. Under pressure from OpenAI and Anthropic, The Decoder observed, Google's former research lab has become a Gemini factory.

Whether that factory can deliver Gemini 4 before its rivals extend their lead is now the central question of Google's AI year.

---

Stay Ahead of AI

Get the latest AI news, analysis, and breakthroughs — all in one place.

Read more AI news →