At the Alphabet Q2 2026 Earnings Call, CEO Sundar Pichai revealed the latest progress on the next-generation cutting-edge large model "Gemini4." In the context of the delayed release of Gemini3.5Pro, which was originally scheduled for June and is still in the partner testing phase, Google has launched the largest pre-training program in its history to fully develop Gemini4, ensuring a leading position in the competition for general artificial intelligence (AGI).

gemini, Google

In terms of technology and product layout, Google has recently released Gemini3.6Flash, 3.5Flash-Lite, and 3.5Flash Cyber for security purposes. To address previous shortcomings in code generation and agent (Agent) capabilities, Google has significantly improved the performance of the 3.x Flash series through iterative optimization. To ensure the training of cutting-edge models, Google has clearly prioritized maintaining a leading position in AGI by allocating tensor processing unit (TPU) computing power, and has built a larger foundational model for Gemini4. Pichai stated that Google's future model delivery rhythm will significantly accelerate, with the goal of achieving an iteration frequency of "releasing a new model almost every month."

Industry experts expect Gemini4 to officially debut between November and December this year. Faced with increasingly fierce AI competition, Google has demonstrated its determination to maintain its position in the forefront of technological innovation by prioritizing core computing power and accelerating the iteration cycle. This also indicates that the development of large models in the industry is rapidly entering a new stage characterized by frequent updates and efficient collaboration.