A preview of Google's next-generation flagship AI model, Gemini 3 Pro, has been spotted on Google's enterprise platform, Vertex AI. The model is tagged "gemini-3-pro-preview-11-2025," and is estimated to be released in November 2025.
Google's most powerful Gemini 3 Pro AI model is rumored to release in November: 1 million tokens of context.
Based on the model tag and existing leaks, Google anticipates opening a preview version of the model to select enterprise customers in November 2025, and possibly releasing it to the public in December of the same year. As Google's flagship AI series, Gemini aims to unify the processing of various information such as text, images, audio, video, and code, and the emergence of Gemini 3 Pro suggests a major revolution in AI interaction.
The most anticipated breakthrough of the new model is its rumored context window of up to 1 million tokens. This capacity, far exceeding current mainstream models, means that the Gemini 3 Pro will possess game-changing capabilities for handling extremely long texts. The context window refers to the length of information an AI model can "remember" and process in a single dialogue or task. One million tokens are equivalent to a 300-page book (approximately 750,000 English words), meaning the model can read and analyze a very thick book at once without forgetting the beginning.
Specifically, it can analyze entire books, complex legal documents, or research papers in one go, deeply understand large codebases, and maintain logical coherence in long, multi-step dialogues. This feature will greatly improve its efficiency and accuracy in tasks such as advanced data analysis and complex information extraction.
Regarding multimodal capabilities, the Gemini 3 Pro will further evolve from its predecessor. It is speculated that it will possess more complex image and video analysis, more accurate speech recognition and generation capabilities, and stronger cross-modal reasoning capabilities, enabling it to better connect and understand the deep relationships between different data types.
In addition, the recently revealed new image generation model from Google, Nano Banana 2, may also be released in conjunction with Gemini 3 Pro. The combination of the two may achieve unprecedented results in fields such as text-to-image and image-to-text, demonstrating Google's strategic intention to build a comprehensive AI ecosystem.
