Google has taken a significant step toward its next generation of artificial intelligence by officially starting the pre-training phase for Gemini 4. At the same time, the company is expanding its Gemini lineup with a new wave of efficient Flash models aimed at developers and enterprise customers.
The update, shared by Tulsee Doshi, underscores Google’s focus on delivering more capable AI models while improving efficiency and reducing inference costs. Alongside confirming that Gemini 4 is now in pre-training, Google also introduced Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber—new models optimized for coding, reasoning, and cybersecurity workloads.
Together, these announcements suggest Google is investing in both its future flagship AI model and a broader ecosystem of specialized, production-ready models.
Gemini 4 Enters Pre-Training
The biggest announcement is that Gemini 4 has officially entered the pre-training stage, marking the beginning of development for Google’s next frontier AI model.
Pre-training is the foundational phase in building a large language model. During this stage, the model is trained on massive datasets to learn language, reasoning, coding, and multimodal capabilities before moving on to instruction tuning, reinforcement learning, and safety testing.
Google has not revealed any technical specifications or a launch timeline for Gemini 4. However, entering pre-training is an important milestone that indicates active progress on what is expected to become the company’s most advanced AI model yet.
The announcement also comes as competition in the AI industry continues to intensify. According to the update, DeepMind employees have expressed grounded optimism about Gemini 4’s potential as Google competes with leading AI companies including OpenAI, Anthropic, and xAI.
While many details remain under wraps, the start of pre-training confirms that Google’s next major AI model is officially in development, setting the stage for the next chapter of the Gemini family.
Industry Perspective & Future Outlook
Google’s latest updates demonstrate a clear shift toward model specialization and token efficiency. By reducing inference costs today with Flash models while quietly training Gemini 4 in the background, Google aims to retain developers looking for pragmatic scalability.
For in-depth analysis on Android, Pixel hardware, and the broader Google AI ecosystem, stay updated with full coverage on NokiaPowerUser.
