Google is unveiling its new artificial intelligence version, Gemini 2.5, at breakneck speed—just a few short months after launching version 2.0. The new model, described by the company as a “thinking model,” is designed to handle more complex tasks, conduct deep analysis, and understand context, rather than simply providing answers based on guesswork or basic classification.
This is a continuation of the Gemini series, first launched last December, now in the form of Gemini 2.5 Pro Experimental—a prototype model that operates fundamentally differently from its predecessors. It focuses less on response speed or efficiency and more on activating a thought process that simulates reflection, analysis, and synthesis of information.
Google emphasizes that this is a deep upgrade not only to the model's core but also to the post-training process. According to Koray Kavukcuoglu, VP of Research at Google DeepMind, the goal now is “to incorporate these thinking capabilities into every future model, so they can handle more complex problems and support intelligent agents that are aware of context and task.”
Among its impressive capabilities, the model supports an enormous amount of information—up to one million tokens, equivalent to about 750,000 words—and will soon be expanded to two million tokens. The implication: It can process extremely long texts, images, code, audio files, and more—while still maintaining consistency and focus.
In the development domain, Gemini 2.5 excels at working with code. It can write entire applications, edit and improve existing code, and even act as an autonomous agent performing complex tasks for developers. At Google, it is referred to as a “code monster,” particularly suited for engineers and professional developers.
Even in objective performance metrics, the model has achieved especially impressive results. Google notes that it broke records in several benchmark tests, including Humanity’s Last Exam—a notoriously difficult test that challenges even the most skilled humans.
Access to the model is already available to Gemini Advanced users through the Gemini mobile app, as well as on the Google AI Studio platform via browser. In the near future, access will also be extended to users of Vertex AI, Google’s cloud service, allowing the model to be integrated into large-scale industrial and commercial tools.
Meanwhile, Google continues to integrate previous Gemini versions—like 2.0—into its various products, including the revamped search engine with AI Mode and deep search tools such as Deep Research. With the new version, Google expects even smarter applications that can assist users with decisions, analyses, and in-depth research.