Google DeepMind head Koray Kavukcuoglu announced that the company is currently developing its next-generation flagship model, Gemini 4. The project has entered the early stages of post-training, a process focused on refining the base model to improve reliability.
Kavukcuoglu stated that the company intends to release the model well before the end of the year. The development team is currently prioritizing safety testing and the implementation of protection mechanisms for the new architecture.
Internal engineers are already testing Gemini 4 by utilizing it to operate the Antigravity AI programming tool. This release will follow the Gemini 3 series, which debuted late last year, and the subsequent Gemini 3.1 update released in February.
Google previously shifted its focus toward smaller, more cost-effective Flash models after the release of Gemini 3.1. While the company had initially planned to launch a Gemini 3.5 Pro version in June, Kavukcuoglu confirmed that the current development priority is now exclusively Gemini 4.
