Analysis
Koray Kavukcuoglu, the new head of Google DeepMind, said in his first public comments since taking the role that Gemini 4 has moved into post-training -- the refinement stage where a base model is tuned to behave reliably -- and that Google wants to release an early version "as soon as possible," according to 9to5Google and Northeast Times, both reporting on September 24. He told The Information that internal results have been strong enough to justify shipping earlier than Google's original before-year-end target.
Value Add Pulse previously covered Sundar Pichai's July 23 announcement that Google had begun what he called its most ambitious pretraining run yet for Gemini 4, at the time framed alongside a delayed Gemini 3.5 Pro release over coding-capability gaps. Two months later, the model has cleared pretraining and moved into post-training -- real forward progress, and the first time a Google DeepMind executive has attached a specific urgency signal to the timeline rather than a general year-end target.
“Google has not shipped a new flagship model since the Gemini 3 series in November 2025.”
The competitive backdrop explains the urgency. Google has not shipped a new flagship model since the Gemini 3 series in November 2025. In the ten months since, OpenAI has released GPT-6 and Anthropic has shipped a new model tier, and both are now generally regarded as ahead of Gemini 3 on public benchmarks and developer mindshare. A new DeepMind chief making ship-speed his first public message, rather than talking up capability gains, is itself informative about what Google thinks its actual problem is.
Post-training is also where frontier-model timelines most often slip -- safety evaluation, red-teaming and alignment tuning routinely add months once a model is otherwise ready, which is why Kavukcuoglu's "as soon as possible" is a direction of travel rather than a date. Google has not said whether an early release would be a limited developer preview or a full general-availability launch, a distinction that matters more to enterprise customers planning around it than the headline timing does.
