Smarter than others. Google releases most powerful AI model Gemini 2.5 Pro
Google's new AI model could be a game changer (photo: Getty Images)
After updating the models available to all users, Google has announced Gemini 2.5 Pro (experimental) for Advanced subscribers and developers, reports 9to5Google.
What is known about the Gemini 2.5 Pro model
A feature of all Gemini 2.5 models, including future versions, is that they are "thinking models", capable of thinking through their answers before they give them. This increases accuracy and productivity.
Google is building these thinking abilities into all of its models so that they can solve more complex problems and support more advanced, contextually aware agents.
Compared to 2.0 Flash Thinking, first introduced in December and updated this month, Google no longer uses explicit "Thinking" labeling. However, in the Gemini app, users can turn on the Show Thought Progress feature to see the model's thought process.
In the field of AI, "reasoning" means not only classification and prediction, but also analyzing information, drawing logical conclusions, taking into account context and nuances, and making informed decisions.
Gemini 2.5 offers a new level of performance thanks to a significantly improved base model and an enhanced post-training process.
The first model in this family is Gemini 2.5 Pro (codenamed "nebula"). It is focused on performing complex tasks, and Google notes that it took first place in the LMArena rating (measuring user preferences) by a wide margin. The model also leads in math (AIME 2025) and science (GPQA diamond) tests.
Additionally, Gemini 2.5 Pro achieved a leading 18.8% score without tools in the Humanity's Last Exam, a dataset created by hundreds of experts to test the limits of human knowledge and logical thinking.
Gemini 2.5 Pro shows an advanced result (photo: 9to5Google)
Google places a special emphasis on advanced programming, claiming that Gemini 2.5 Pro is "a huge step up from 2.0" and that "further improvements are already in the works."
The model does an excellent job of creating visually appealing web applications and "agent" code, as well as transforming and editing code.
On the SWE-Bench Verified test, which is an industry standard for evaluating standalone code agents, Gemini 2.5 Pro scored 63.8% when using a custom agent configuration.
In addition to the built-in multimodality, Gemini 2.5 Pro has a context window of 1 million tokens, and this will soon increase to 2 million.
It can process vast amounts of data and handle complex tasks using various sources of information - text, audio, images, video, and even entire code repositories.
In the app itself, Gemini 2.5 Pro will replace the experimental 2.0 Pro version (photo: 9to5Google)
Gemini 2.5 Pro (the experimental version) will first be available to Gemini Advanced subscribers and Google AI Studio users, and in a few weeks it will be available on the Vertex AI platform.
In the near future, Google will also introduce 2.5 Pro pricing, which will allow the model to be used with higher query limits for large-scale industrial use.
In the Gemini app, it will replace the experimental 2.0 Pro version, adding support for Gmail, YouTube, and file uploads.
You may be interested in:
- How to build AI bots with Google Gemini with a simple guide
