Google has been watching rivals ship major model updates while it sat quietly with Gemini 3, which launched back in November 2025. That gap is now closing, at least according to the man now running DeepMind. In his first media appearance since taking over the division, Koray Kavukcuoglu told The Information that Gemini 4 is in its final refinement stage and that Google plans to release it “much earlier” than the end of 2026.
“Our intention is to, like, as soon as possible, release an early post-training output because we see the results and we are excited,” Kavukcuoglu said, adding that Google would “continue the fast-paced iterations” going forward. That’s a noticeably candid tone from a new leader who clearly wants to signal momentum. But the context here matters a lot.
Since Gemini 3 shipped, OpenAI released GPT-6 and Anthropic put out its Mythos series. Both product lines outperform Gemini 3 on standard benchmarks and have given those companies a meaningful lead in the eyes of developers, enterprise buyers, and the broader market. Google’s absence at the flagship model level has been felt. Sundar Pichai promised a Gemini 3.5 Pro update at I/O in May, targeting June availability, but that release never happened. So when Kavukcuoglu talks about fast-paced iterations, there’s some ground to reclaim before that framing lands with much credibility.
Kavukcuoglu acknowledged the delay directly, saying Google “took a little bit of a step back” and focused on shipping faster, less powerful Flash models rather than pushing out Gemini 3.5. Whether that was the right strategic call is debatable. Flash models have practical utility, especially for cost-sensitive developers who don’t need frontier-level reasoning. But skipping the mid-cycle update while competitors shipped headline models created a perception problem that Google is now trying to correct.
The leadership transition at DeepMind adds another layer here. Former AI chief Demis Hassabis stepped down in August after signaling that “great progress” was being made on Gemini 4. Replacing a founder-level figure mid-cycle is never a clean handoff, and Kavukcuoglu’s early public appearances are clearly designed to reassure the market that nothing is broken internally. The optimistic framing around Gemini 4’s readiness fits that narrative.
For developers evaluating their model stack, the timing of Gemini 4’s release will matter more than the announcement. Google has strong infrastructure advantages, deep integration across Workspace and Cloud, and competitive pricing at the API level. But if Gemini 4 ships and lands below GPT-6 or Mythos on reasoning and coding benchmarks, the gap becomes harder to explain away with roadmap promises.
The AI model market in 2026 is not especially forgiving of slow release cadences. Anthropic and OpenAI have both demonstrated that consistent, incremental shipping builds trust with technical users faster than big-bang releases. Google knows this. Gemini 4 being described as nearly ready is genuinely promising, but “much earlier than end of year” is still a wide window. The details of what the model actually does will matter far more than the announcement that it’s coming.
