Google is closing in on the launch of Gemini 4, the flagship artificial intelligence model it has spent much of the past year promising, according to the executive now running its DeepMind division.
Koray Kavukcuoglu, who took charge of DeepMind following a leadership reshuffle over the summer, used his first media appearance in the role to say the model has entered its refinement stage and that the company intends to ship it considerably sooner than the end of 2026. He made the comments to The Information, giving the clearest public timeline yet for a release Google has left conspicuously undefined while rivals moved ahead.
Gemini 4 Is Being Refined, Not Rebuilt
Kavukcuoglu framed the current work as polishing rather than a fundamental rethink of the underlying approach. He said Google's intention is to release an early post-training output as soon as possible, because internal results have been encouraging enough to justify moving quickly. He added that the company expects to continue the fast-paced iterations that follow a major release.
That phrasing carries real implications. Frontier models typically pass through pre-training, post-training and safety evaluation before they reach the public, and the point at which a lab chooses to ship depends heavily on how much further refinement it believes is worth waiting for. By describing an early post-training release, Kavukcuoglu signals a willingness to put a version of Gemini 4 into circulation and improve it through subsequent updates, rather than holding the entire program back until every capability is finalised.
It is a pattern that has become familiar across the industry, where flagship launches are increasingly treated as starting points rather than finished products. For Google, it would also mark a return to a more predictable rhythm of releases after an unusually quiet stretch.
A Year Without a Flagship Release
The pressure behind the Gemini 4 timeline is easy to understand. Google has not shipped a new flagship model since the Gemini 3 series arrived in November 2025, leaving a gap of nearly a year at the top of its lineup. In that window, two major competitors have released new generations:
- OpenAI's GPT-6, which has arrived since Gemini 3 and performs better than Google's current flagship.
- Anthropic's Mythos series, which has followed a similar path and also outperforms Gemini 3.
Together, those launches have handed rival developers a meaningful advantage in the race for frontier performance. For an organisation that has publicly positioned itself at the leading edge of AI research, watching two competitors claim the top of the benchmark tables is an uncomfortable position — and one that has made each new signal about Gemini 4 more closely scrutinised.
Leadership Change at the Top of DeepMind
The Gemini 4 timeline has unfolded against significant churn inside Google's AI organisation. Demis Hassabis stepped down from his role as head of AI in August, departing with a public statement that great progress was being made on Gemini 4. Kavukcuoglu's elevation to lead DeepMind followed, making his first round of media interviews a closely watched introduction.
Leadership transitions at research labs rarely stay contained to org charts. New leaders often recalibrate priorities, adjust release schedules and shift how much the company says publicly about work in progress. Kavukcuoglu's decision to speak openly about both the state of Gemini 4 and the reasoning behind earlier delays suggests a deliberate attempt to reset expectations — and to demonstrate that the division has a plan it is prepared to be measured against.
The Gemini 3.5 Pro That Never Shipped
Part of what made the silence around Google's flagship line so noticeable was that the company itself kept setting dates. At the I/O event in May, chief executive Sundar Pichai said a major Gemini 3.5 Pro update would arrive in June, positioned explicitly as a way to bridge the performance gap with competing models. That release never happened.
The missed milestone became the clearest illustration of the distance between Google's public roadmap and what its teams ultimately delivered. It also raised questions about whether the company was struggling to keep pace, or whether it had simply decided that shipping an interim flagship was not the right use of resources. Kavukcuoglu's account points toward the latter explanation.
Flash Models and a Deliberate Step Back
According to Kavukcuoglu, Google took a little bit of a step back in order to concentrate on its Flash models — faster, lighter systems that prioritise speed and cost efficiency over raw capability — instead of pushing out Gemini 3.5 Pro.
That is a defensible strategic choice, even if it looks like a retreat from the outside. Lighter models are cheaper to run at scale, respond more quickly and are easier to deploy across consumer products used by enormous numbers of people. Building out that tier can generate immediate practical value while a larger, more expensive flagship is still being tuned.
The trade-off, however, is visibility. Flash releases do not generate the same headlines or benchmark comparisons as a new frontier model, which meant Google spent months appearing absent from the conversation even while its engineering teams were shipping.
Confidence About Staying at the Frontier
When asked directly whether he was concerned that Google was falling behind its competitors in AI development, Kavukcuoglu was unambiguous. He said that in his mind it is a certainty that the company will always be at the frontier.
That confidence will ultimately be tested by results rather than statements. The gap created by Gemini 3.5 Pro's absence and Gemini 3's age is measurable in benchmark performance, and closing it requires a model that can be compared directly against GPT-6 and the Mythos line. Kavukcuoglu's promise of continued fast-paced iterations is, in effect, a commitment to a schedule — one that will be judged release by release.
What to Watch For Next
Several open questions will shape how the Gemini 4 launch is received:
- Whether the model actually arrives well before the end of 2026, as Kavukcuoglu has indicated, or slips past that window in the way Gemini 3.5 Pro did.
- Whether it closes the performance gap with GPT-6 and Anthropic's Mythos series, or merely narrows it.
- Whether Google re-establishes a consistent cadence of flagship releases, or treats Gemini 4 as a one-off correction.
- How the Flash line and any future interim updates fit alongside the flagship tier in Google's product strategy.
For now, the clearest takeaway is that Google has moved from silence to a stated timeline. Gemini 4 is being refined, its leadership says the results are exciting, and the company wants it out the door quickly. After nearly a year on the sidelines of the flagship race, the next few months will determine whether that urgency translates into a genuine return to form.
This article is based on reporting by The Verge. Read the original article.
Originally published on theverge.com
