DeepMind's new chief says Gemini 4 is in refinement and could ship well before year end
By EnkiReporting from The Verge
Koray Kavukcuoglu told The Information that Google aims to release an early post-training version of Gemini 4 as soon as possible, after nearly a year without a new flagship model while OpenAI and Anthropic moved ahead.
Google may be close to releasing Gemini 4, its first new flagship model in close to a year. Koray Kavukcuoglu, speaking to The Information in his first media interview as head of Google DeepMind, said the model is being refined and that the company wants it out "much earlier" than the end of 2026, The Verge reported.
Kavukcuoglu said Google intends to ship an early post-training version as soon as it can because the team is excited by the results, and that faster iteration would follow.
A long gap at the top
Google's most recent flagship release was the Gemini 3 series in November 2025. In the months since, OpenAI has launched GPT-6 and Anthropic has released its Mythos series, and The Verge describes both as outperforming Gemini 3, leaving Google playing catch-up.
The company had tried to close that gap earlier. At its I/O conference in May, CEO Sundar Pichai said a major Gemini 3.5 Pro update would arrive in June. It never shipped. Kavukcuoglu said Google stepped back from that release to prioritize its faster, less capable Flash models instead.
Leadership change at DeepMind
The interview also marks a new public face for Google's AI effort. Demis Hassabis stepped down as DeepMind's leader in August, after saying the lab was making strong progress on Gemini 4. Kavukcuoglu now carries the message, and he rejected the idea that Google is falling behind, telling The Information he considers it certain that Google will remain at the frontier.
No release date, benchmarks or pricing for Gemini 4 were given, and the timeline remains a stated intention rather than a commitment. Google's recent record, with a promised June model that never appeared, is a reason to treat the target cautiously.
Why the timing matters
For the model market, a credible Gemini 4 would restore three-way competition at the top end, which has leaned toward OpenAI and Anthropic since late 2025. In the meantime, Google chose to put its effort into Flash models that favor speed over peak capability.
An early post-training release, as Kavukcuoglu described it, suggests Google may put a model in developers' hands before it is fully polished and improve it through quick updates. That approach can deliver gains sooner but also means behavior may shift between versions.
What it means for founders
- Keep model choice flexible. If Gemini 4 lands this year, a third top-tier option could change price and performance comparisons, so avoid hard-coding a single provider into core workflows.
- Plan for moving targets. An early release followed by rapid iteration can mean changing outputs; pin versions where possible and keep evaluation sets ready to rerun.
- Flash models are Google's current bet. For latency-sensitive or high-volume products, Google's cheaper, faster tier is where it has been investing while the flagship waits.
- Discount dates until they ship. The missed Gemini 3.5 Pro launch is a reminder not to schedule product launches around an unreleased model.
Sources
Enki Daily
Get stories like this every weekday morning.
The day's AI stories for founders, each with what it means for your company. Free.
Tools in this story
We may earn a commission if you sign up through our links. It never affects our ratings or which stories we cover.
Google's assistant, built into Workspace and Android