Google DeepMind is nearing the launch of its next flagship AI model, Gemini 4. Koray Kavukcuoglu, head of the division, stated that the model is currently in the early stages of post-training, a process focused on refining performance and reliability. The company aims to release the model well before the end of the year, with plans for rapid iterations following the initial launch.
The development of Gemini 4 marks a shift in strategy for Google, which previously prioritized the release of smaller, cost-effective Flash models over the anticipated Gemini 3.5 Pro update. While Google did not explicitly confirm the status of the 3.5 Pro project, the company is now fully focused on its next-generation flagship. Engineers are currently testing Gemini 4 internally using the Antigravity AI programming tool.
Google faces stiff competition from Anthropic and OpenAI, which have recently released advanced models like Mythos and GPT-6. Despite this, Kavukcuoglu expressed confidence in his team's ability to remain at the forefront of AI technology. He emphasized that the primary goal is building trustworthy AI agents rather than focusing on the debate surrounding the achievement of Artificial General Intelligence.
The advancements in Gemini 4 are also expected to influence Google's hardware division, specifically the development of TPU AI chips. Google has begun selling these chips directly to customers to compete with Nvidia. Research from the DeepMind team provides critical data for hardware engineers, helping them plan the design architecture for the next two to three generations of Google's proprietary AI processors.
