Gemini 4 Coming Soon as DeepMind Sets AGI Debate Aside

Google DeepMind's new leader has a clear priority, and it is not artificial general intelligence. Koray Kavukcuoglu wants Gemini 4 out well before the end of the year. He has also made it plain that he sees little value in arguing over whether AGI has already arrived.

Kavukcuoglu made his first media appearance as head of Google DeepMind at The Information's AI Agenda Live Summit. There, he described Gemini 4 as being in an early phase of post-training, the stage where a base model is refined so that it behaves reliably. He said he hopes to release it considerably earlier than year-end.

The plan is to ship an early post-trained version as quickly as possible, because the team likes what it is seeing, and then iterate fast. Safeguards and safety testing are still underway. Google engineers are already using the model internally with Antigravity, the company's coding tool.

Kavukcuoglu, previously DeepMind's CTO, took charge in August after Demis Hassabis stepped down as CEO. Hassabis now chairs the unit and serves as Alphabet's Chief Scientist.

A gap to close

Gemini 4 would be Google's first new flagship since Gemini 3 arrived late last year. Since then, the company has shipped the Gemini 3.1 update in February and several smaller, cheaper Flash models. Gemini 3.5 Pro, which Sundar Pichai announced at I/O in May for a June release, never appeared. According to the Financial Times, the project missed three deadlines.

Kavukcuoglu described the move as taking a small step back to maximise the speed of learning. He did not say whether Gemini 3.5 Pro has been formally cancelled.

Rivals have not waited. Anthropic launched its Mythos/Fable family in the spring and updated both that line and its Opus series this month. OpenAI has been releasing GPT-6 models since early September. All of them are significantly more capable than Google's strongest current models.

Kavukcuoglu says he is not worried. He called himself completely confident, praised what he described as an exceptionally strong team, and said it is a certainty that Google will always be at the top.

The model work also feeds into hardware. Since this year, Google has sold its TPUs directly to customers, putting it in more direct competition with Nvidia. According to Kavukcuoglu, new models show the chip engineers how the next two or three generations of TPUs should be designed.

Google has closed a gap before. Gemini 3 was a big enough leap that OpenAI declared a "Code Red" and only caught up with GPT-5.2. Whether Gemini 4 can repeat that remains an open question.

From AGI mission to product engine

What Kavukcuoglu said about AGI may matter more than his timeline. Whether the industry has reached AGI, he said, is "not the right discussion". The real question is whether it can build intelligent agents that people can trust, an issue closely tied to where the security boundary of AI agents sits.

That is a notable shift for a lab that long defined itself by the pursuit of general intelligence. Only weeks ago, Hassabis wrote in his farewell memo that he had worked towards AGI his whole life and now considered it within reach.

Hassabis co-founded DeepMind in 2010, and Google acquired it four years later. The London lab long held a special position inside the company. ChatGPT changed that. During Google's first "code red", Google Brain and DeepMind were merged into Google DeepMind, which is also where the Gemini name comes from.

August's restructuring goes further. Kavukcuoglu runs day-to-day operations without a CEO title, and all Gemini development is moving to the Bay Area. A former Google manager told the Guardian that DeepMind's era as an independent player is over. The Financial Times reported that executives were frustrated that Hassabis gave commercial needs less weight, for example in the decision to release AlphaFold openly.

How voluntary Hassabis' exit was is disputed. The newsletter Pathfounders, citing industry sources, reported that he wanted to leave together with Jeff Dean, but management feared a drop in the share price. Google has not commented.

The talent drain is not in dispute. Noam Shazeer joined OpenAI in June, and two days later AlphaFold Nobel laureate John Jumper announced his move to Anthropic. Dean founded the start-up Discovery Loop with Oriol Vinyals, Quoc Le and Sanjay Ghemawat, while Gemini researchers Jonas Adler and Alexander Pritzel also went to Anthropic. According to Fortune, DeepMind's ratio of hires to departures fell from about 12 to 1 in Q2 2023 to roughly 2 to 1 in Q3 2026.

Timing that stands out

The AGI question is unusually prominent right now. On 12 September, Anthropic CEO Dario Amodei argued in his essay "We Must Pace the Frontier" that leading labs should increase model capabilities more slowly without halting development. OpenAI followed, Elon Musk agreed, and Hassabis called the direction right while saying the details still need work. In July, more than 1,200 frontier-lab employees had signed a letter of the same name, part of a wider debate that also includes calls for pausing frontier models.

Kavukcuoglu is not dismissing safety, since trustworthy agents are a safety issue in their own right. But while others debate AGI and its risks, he treats progress as a product question. The pressure from OpenAI and, increasingly, Anthropic has turned a research lab with an AGI mission into Google's Gemini factory, and that transformation now looks complete.