Google DeepMind's new head, Koray Kavukcuoglu, said Gemini 4 is in the early phase of post-training and that the company aims to release the flagship model "much earlier" than the end of the year1,2. The comments were made during Kavukcuoglu's first media appearance as leader of Google DeepMind.
"Our intention is to, like — as soon as possible — to release an early post-training output because we see the results and we are excited," Kavukcuoglu told The Information. He added that Google would "continue the fast-paced iteration" after the initial release.
Kavukcuoglu described Gemini 4 as being in its "refinement stage," indicating that pre-training is complete and the model is now undergoing the alignment and capability tuning that precede a public launch. The phrasing "early post-training output" suggests Google may ship an initial checkpoint before the full post-training pipeline is finished, then iterate publicly.
The Verge characterized Google as having "dawdled behind rival developers on flagship AI releases". ANALYSIS Kavukcuoglu's emphasis on speed — "as soon as possible" — and his commitment to continued rapid iteration after launch read as a direct response to that gap, framing Gemini 4's rollout as an attempt to close the release-cadence distance with competitors.
The disclosure arrives shortly after Vector Wire's prior coverage of Anthropic's Claude autonomously discovering a novel enzyme system in bacteriophage DNA, the first result from Anthropic's wet biology lab[1]. ◆ The near-simultaneous announcements from two of Google DeepMind's principal rivals underscore the pace at which frontier labs are racing to demonstrate capability milestones.