Google has just unveiled a new AI model called Gemini 2.5 Pro, which the company claims is its most advanced achievement in the Gemini series. The model is designed to perform complex tasks and uses powerful reasoning capabilities that Google calls “Thinking Built-in.” Those using the Gemini Advanced service can start working with this model immediately.
According to Google, the Gemini 2.5 series has a natural reasoning capability. This feature allows the model to examine data, reach logical conclusions, and solve complex challenges with a better understanding of the context of the problems. This ability, called “Thinking Built-in,” is present in all models in this series, although “Thinking” is no longer used in their names. Users can watch the process unfold by enabling the Show Thinking option in the Gemini app.
The Gemini 2.5 Pro, codenamed Nebula and officially named gemini-2.5-pro-exp-03-25, is the first member of the 2.5 series and represents a significant improvement over previous models. Google announced that the model had reached the top spot in the LMArena rankings, which are based on human opinions about the quality of answers.
Outstanding Performance on Specialized Tests
5 Top Chinese AI You Should Know
The model also performed well on two specialized tests. On the AIME 2025, which includes tough math questions at the Olympiad level, and on the GPQA diamond, which measures the ability to reason scientifically and answer complex questions, the Gemini 2.5 Pro scored the highest. Notably, this success was achieved without using “majority voting,” in which multiple answers are generated, and the most frequently chosen ones are selected.
On Humanity’s Last Exam, the model also achieved a score of 18.8 percent without the help of external tools. Google considers this result to be the best performance recorded among models measured solely on their internal abilities. The benchmark, which assesses knowledge and reasoning in more than 100 subjects, presents a detailed and diverse set of questions from the humanities, natural sciences, and analytical sciences.
The improvements in Gemini 2.5 Pro result from fundamental changes to the core model architecture and enhancements to the post-development training phase. The model is built to solve more complex problems and support agents that require a deeper understanding of text and context. The model also has a breakthrough in programming. Google says Gemini 2.5 Pro has better coding capabilities than version 2.0 and delivers reliable performance in building web apps, designing agent-based coding tools, and editing code. On the SWE-Bench Verified benchmark, which measures automated code generation, the model achieved a score of 63.8 percent with the help of a dedicated agent.
Expanding Data Processing Capabilities
Google’s new model supports a context window of 1 million tokens and will soon expand to 2 million. This allows the model to simultaneously process large amounts of data, such as entire code repositories. In addition, Gemini 2.5 Pro can simultaneously analyze data in multiple formats, including text, audio, images, and video.
The model is available through Gemini Advanced and Google AI Studio and will be added to Vertex AI, Google’s cloud platform for developing enterprise AI models, in the coming weeks. Google will soon announce more information about pricing and capacity options for larger projects.
FAQ
1. What new features does the Gemini 2.5 Pro model have?
The Gemini 2.5 Pro model has advanced features, including natural reasoning capabilities that allow the model to analyze data, reach logical conclusions, and solve complex challenges with a better understanding of the context. This feature, called “Thinking Built-in,” is available to users.
2. What are the differences between the Gemini 2.5 Pro model and its previous versions?
The Gemini 2.5 Pro model has several improvements in its core architecture and post-development training compared to previous versions. It can solve more complex problems and performs better in programming, web app development, and code editing than version 2.0.
3. Which tests demonstrate the superiority of the Gemini 2.5 Pro model?
The Gemini 2.5 Pro model excelled in two specialized tests. It scored the highest in the AIME 2025 exam, which includes Olympiad-level math questions. It also ranked highly in the GPQA test, which measures the model’s scientific reasoning ability and complex question-answering performance.
4. How well does Gemini 2.5 Pro perform in coding?
The Gemini 2.5 Pro model outperforms version 2.0 in coding. It is reliable in building web apps, designing agent-based coding tools, and editing code. In the SWE-Bench Verified benchmark, which assesses automated code generation, it achieved a score of 63.8 percent.
5. What are the practical applications of the “Thinking Built-in” feature?
The “Thinking Built-in” feature enables the model to analyze data, make logical deductions, and intelligently solve complex problems. It is beneficial in fields requiring precise reasoning and a deep understanding of context.
6. How can we use the Gemini 2.5 Pro model?
The Gemini 2.5 Pro model is available through the Gemini Advanced service and Google AI Studio. In the coming weeks, it will also be integrated into the Vertex AI platform, which is designed for developing enterprise AI models.
7. What data processing capabilities does Gemini 2.5 Pro have?
The Gemini 2.5 Pro model can process large volumes of data, supporting a context window of 1 million tokens. It can simultaneously analyze data in multiple formats, including text, audio, images, and video. This capacity will soon be expanded to 2 million tokens.