DeepMind Chief Sets Sights on Speeding Up Gemini 4 Launch
Koray Kavukcuoglu, the new head of Google DeepMind, made his first public remarks in the job and surprised many by saying that Gemini 4 has entered early post-training. This could potentially lead to a much earlier launch than the previously assumed late-2026 window.
Gemini 4's pre-training run started as announced on July 21, when Sundar Pichai confirmed the project. Since then, Google wants to move towards an almost monthly release cadence for Gemini, which is a significant shift from its historically slower pace.
The urgency behind Gemini's launch is due to the model falling behind in coding workloads, with GPT-5.5 and Anthropic's Claude Code being the current defaults. Kavukcuoglu stated that Google wants to roll out an early post-training version as soon as possible because they've seen promising results.
Alphabet's stock wobbled on the news, but recovered slightly after Kavukcuoglu's remarks. The practical question for developers and enterprise buyers is whether an earlier Gemini 4 launch will change what they build on right now. Post-training can take weeks or months depending on how much safety testing and fine-tuning Google wants to do before a public launch.