Google's flagship Gemini 4 model has reached its refinement stage and the company is aiming to release it well before the end of 2026, according to Koray Kavukcuoglu, who spoke to The Information in his first media appearance as head of Google DeepMind. He said the intention is to put out an early post-training output as soon as possible rather than hold the model back, because internal results are encouraging.
Key takeaways
- Kavukcuoglu said Gemini 4 is in refinement and targeted for launch much earlier than the end of the year, with an early post-training checkpoint released first.
- Google has not shipped a new flagship since the Gemini 3 series in November 2025, while OpenAI's GPT-6 and Anthropic's Mythos series arrived in the interim and both outperform Gemini 3.
- The promised Gemini 3.5 Pro update, announced for June at I/O, was never released; Kavukcuoglu said the team stepped back to concentrate on Flash models instead.
A ten-month gap at the top of the lineup
The context that makes the comment newsworthy is the silence preceding it. Google's last flagship large language model release was the Gemini 3 series in November 2025. Sundar Pichai told the I/O audience in May that a major Gemini 3.5 Pro update would land in June. It never shipped.
Kavukcuoglu's explanation, as reported by The Information, is that Google deliberately stepped back from that release to concentrate on smaller, faster Flash-tier models. That is a defensible engineering call and also an accurate description of what Google actually shipped: a run of cheap, high-throughput models rather than a frontier push. Gemini 3.7 Flash arrived three weeks after 3.6 at half the price, and the pattern continued from there.
What shipping an early post-training checkpoint means
The specific commitment is more unusual than the timeline. Releasing an early post-training output means publishing a model whose alignment and instruction-following passes are not finished, then iterating in public. Google has done versions of this before with preview tiers, but framing it as the intended launch path for a flagship is a change in posture โ and an implicit admission that being present in the market now matters more than arriving polished.
Kavukcuoglu also said Google would continue fast-paced iterations after the initial release, which reads as a commitment to a shorter cadence than the roughly annual flagship rhythm Google has kept.
The leadership change behind it
Kavukcuoglu took over after Demis Hassabis stepped down as Google's AI head in August, having said on the way out that great progress was being made on Gemini 4. A new leader using a first interview to commit to a date is a conventional reset move, but it does convert a vague internal assurance into a public deadline that can be missed visibly.
Asked whether he worried about Google trailing its rivals, Kavukcuoglu said he regards it as a certainty that the company will always be at the frontier. That is a statement of intent rather than a claim about current benchmark standing, where Gemini 3 sits behind both GPT-6 and the Mythos models.
Context
None of this has stopped Google DeepMind from shipping product. The company made Live Avatar generally available in Gemini Enterprise this week and has been pushing audio and agent tooling steadily. What is missing is the headline model that anchors a lineup and sets the comparison point reviewers reach for.
Outlook
Treat the timeline as a target, not a date. The useful signal is the delivery mechanism: if Google ships an unfinished post-training checkpoint under the Gemini 4 name, expect early benchmark numbers that move week to week and a messier evaluation picture than a clean launch would produce. For developers, the practical question is whether the first checkpoint arrives with stable pricing and API surface, or whether it lands as a preview that changes underneath production code.
FAQ
When will Gemini 4 launch?
Google has not given a date. Kavukcuoglu said the model is in its refinement stage and that the company is aiming for a launch much earlier than the end of 2026, starting with an early post-training release.
Why did Google skip Gemini 3.5 Pro?
Sundar Pichai announced a major Gemini 3.5 Pro update at I/O in May, scheduled for June, but it was never rolled out. Kavukcuoglu said the team took a step back from it to focus on faster, less powerful Flash-tier models instead.
Is Gemini 3 currently behind its competitors?
Yes. OpenAI's GPT-6 and Anthropic's Mythos series both launched after the Gemini 3 series and both outperform it, which is the gap Gemini 4 is meant to close.
