Skip to main content
Google DeepMind CEO Demis Hassabis discusses Gemini 4 AI model, expected before 2026, at a tech conference.

Editorial illustration for Google DeepMind Chief Says Gemini 4 AI Model Arriving Before 2026

Google DeepMind: Gemini 4 Arriving Before 2026

4 min read

Google DeepMind has a new person answering questions about Gemini 4, and his first answer is that it's close. Koray Kavukcuoglu, who took over as DeepMind chief after Demis Hassabis stepped down in August, gave his first interview in the role to The Information. The topic was overdue: Google hasn't shipped a flagship model since the Gemini 3 series landed in November 2025, and the gap has cost the company ground. OpenAI's GPT-6 and Anthropic's Mythos line both arrived in the meantime, and by most measures both now beat Gemini 3 outright.

That's the backdrop for Kavukcuoglu's comments on where Gemini 4 stands and how fast Google wants to move once it's out the door. Hassabis had already flagged "great progress" on the model before his exit, but his departure left an open question about whether Google's release schedule would tighten up or slip further. Sundar Pichai added his own marker back in May, promising a major Gemini 3.5 Pro update at I/O. What Kavukcuoglu told The Information about timing, and about Google's plans to keep iterating quickly afterward, follows below.

Google is reportedly nearing the launch of its long awaited Gemini 4 model, after dawdling behind rival developers on flagship AI releases. Speaking with The Information during his first media appearance as the leader of Google’s DeepMind division, Koray Kavukcuoglu said that Gemini 4 is currently in its refinement stage, and that the company is aiming to launch the model “much earlier” than the end of the year.

Why this matters

Kavukcuoglu's comments are the clearest signal yet that Google feels pressure to close the gap with OpenAI and Anthropic on release cadence, not just benchmark scores. "As soon as possible" and "early post-training output" is notably hedgier language than the polished launch events Google usually stages for Gemini, which tells us DeepMind may ship something rougher than the version they'd normally show off. For developers building on Gemini, that's worth watching closely: an early or iterative release means API behavior, pricing, and capabilities could shift fast in the weeks after launch, not settle into a stable target the way past major versions did.

For founders betting product roadmaps on Gemini 4's rumored capabilities, the "much earlier than end of 2026" framing is still vague enough that we'd treat it as a soft target rather than a shipping date. Anthropic and OpenAI have both burned people with slipped timelines before. The real test isn't when Gemini 4 drops, it's whether Google's first public build looks finished or feels like a checkpoint release rushed out to change the narrative.

Common Questions Answered

When is Google DeepMind planning to launch Gemini 4?

According to Koray Kavukcuoglu, the new DeepMind chief, Gemini 4 is currently in its refinement stage and Google is aiming to launch the model "much earlier" than the end of the year, meaning before 2026. This represents Google's effort to close the gap with competitors like OpenAI's GPT-6 and Anthropic's Mythos line, which have both launched since Gemini 3 arrived in November 2025.

Who is Koray Kavukcuoglu and what is his role at Google DeepMind?

Koray Kavukcuoglu took over as DeepMind chief after Demis Hassabis stepped down in August. He gave his first interview in this leadership role to The Information, where he discussed the upcoming Gemini 4 model and Google's plans to accelerate its flagship AI releases.

Why has Google's delay in releasing Gemini 4 become a competitive concern?

Google hasn't shipped a flagship model since the Gemini 3 series landed in November 2025, and during this gap, both OpenAI's GPT-6 and Anthropic's Mythos line have arrived. This delay has cost Google ground in the competitive AI landscape and signals pressure to close the gap with rivals on release cadence, not just benchmark scores.

What does Kavukcuoglu's language about Gemini 4's launch suggest about the model's readiness?

The use of hedging language like "as soon as possible" and "early post-training output" is notably less polished than the language Google typically uses for Gemini launches. This suggests that DeepMind may ship Gemini 4 in a rougher state than the refined version they would normally showcase at formal launch events.

LIVE14:19OpenAI Agent Breached Australian Health Service, Revealed Months Later