Google prepares to launch Gemini 4 as core training is completed

Google has announced that work on its new flagship artificial intelligence model, Gemini 4, is nearing its final stages. Koray Kavukçuoğlu, who began leading Google DeepMind in August, said the model's core training process has been completed and it has now entered the post-training development stage.

Kavukçuoğlu shared details about Gemini 4 at The Information's AI Agenda Live Summit. According to him, work is currently underway to improve the model's performance, reliability and overall capabilities.

The head of Google DeepMind said the team is pleased with the results achieved so far and that rapid improvements to Gemini 4 will continue. Although the company has not announced a specific launch date, it aims to release the new model well before the end of the year.

Google previously introduced the Gemini 3 family in November 2025. The company later focused on developing smaller and faster Flash models. The source reports that a major update planned for Gemini 3.5 Pro was not released on the expected date.

Google CEO Sundar Pichai announced at the I/O event in May that a major update for Gemini 3.5 Pro would be introduced in June. However, according to the source, this version was ultimately not released to users.

Koray Kavukçuoğlu said Google partially changed its strategy at the time, prioritizing the development of faster Flash models. The company is now primarily focused on developing its next flagship model, Gemini 4.