Gemini 4 almost ready, DeepMind chief says launch 'much earlier' than expected

Gemini 4 is almost ready, according to the new leader of Google DeepMind. In his first media appearance since taking the helm, Koray Kavukcuoglu told The Information that the long-awaited model is in its refinement stage and that Google aims to launch it "much earlier" than the end of the year. He said the company's intention is to release "as soon as possible" an early post-training output because "we see the results and we are excited."

The timing matters. Google has been perceived as dawdling behind rival AI developers on flagship releases, even as its research division continues to publish cutting-edge papers. Kavukcuoglu's comments signal a deliberate acceleration, with the new chief making his debut by telegraphing urgency. For businesses that build on top of frontier models, the prospect of a Gemini 4 launch in the coming months, rather than later, reshapes planning horizons.

Post-training is the final phase where a model is fine-tuned with human feedback and safety measures before public release. An early post-training output suggests Google is willing to ship a version sooner, perhaps as a preview, to gather usage data and iterate quickly. Kavukcuoglu said Google would "continue the fast-paced iteration," underscoring a shift from Google's historically cautious rollout approach.

The context is a hyper-competitive AI market where timing can determine mindshare and enterprise adoption. Google's Gemini series has been well-regarded, but the company has faced criticism for being slower to market than some rivals. A faster release of Gemini 4 could help Google regain momentum, especially among developers who are evaluating which model to standardize on.

For executives, this news is a signal to watch for early access programs or API availability. If Gemini 4 ships ahead of schedule, it could change the calculus for companies currently locked into other model providers. The "early post-training output" might be a beta or a research preview, but it still offers a chance to test the model's capabilities ahead of a full launch.

Kavukcuoglu's leadership is also a factor. He has been at DeepMind for years and his promotion to the top job signals continuity with a research-first culture, but his public comments now indicate a more commercially aggressive stance. That could mean more resources are being diverted to productization and speed.

The stakes are high. Google's cloud business and its AI offerings are a key growth driver, and falling behind in the model race could cost it enterprise contracts. A quick Gemini 4 launch is a direct answer to competitors who have been shipping newer models throughout 2024. For decision-makers, the message is clear: the AI landscape is about to get more crowded, and the window to adapt is shrinking.