Google cooks up its ‘most intelligent’ AI model to date

GEMINI 2.5 is welcomed by Google DeepMind as “the most intelligent artificial intelligence model” so far.

The first model of this last generation is a trump 2.5 Pro experimental version, which DeepMind says has achieved newer results through a wide range of standards.

According to Koray Kavukcuoglu, CTO OF Google DeepMind, Gemini 2.5 models are “thinking models”. This indicates their ability to think through their ideas before generating a response, which leads to promoting performance and improving accuracy.

Kavukcuoglu explains that the ability to “think” goes beyond mere classification and prediction. It includes the ability of the system to analyze information, conclude logical conclusions, integrate context and nuances, and eventually make enlightened decisions.

Deepmind explores methods to enhance the capabilities of intelligence and think about artificial intelligence for some time, as it employs techniques such as reinforcement learning and an idea chain. This basis led to the last input of the first thinking model, Gemini 2.0 Flash Thinking.

“Now, with Gemini 2.5, we have achieved a new level of performance by combining a largely enhanced basic model with the improvement of training,” says Kavukcuoglu.

Google plans to integrate these thinking capabilities directly into all its future models-which leads to the treatment of more complex problems and support for more capable agents.

Gemini 2.5 Pro secures the first place in LMARNA

Gemini 2.5 Pro experimental as the most advanced DeepMind model is developed to deal with complex tasks. As of the writing of this report, I got the first place on the LMARNA-which is a major measure for assessing human preferences-a large margin, which indicates a very capable model in high-quality style:

Gemini 2.5 is a “professional” in mathematics, science, coding and thinking

GEMINI 2.5 Pro has shown the latest model across the various criteria that require advanced thinking.

It is worth noting that it leads to mathematical standards and science-such as GPQA and AIME 2025-without relying on test time techniques that increase costs, such as majority vote. It also achieved the latest degree of 18.8 % in the last humanity exam, which is a collection of data designed by the subject experts to assess the human limits of knowledge and thinking.

Deepmind has focused heavily on coding performance, and GEMINI 2.5 represents a big jump forward compared to its predecessor, 2.0, with more improvements in the pipeline. 2.5 Pro excels in creating visually convincing web applications and software applications, as well as converting and editing software.

On Swe-Bused, achieve the industry standard for the valuors of the agent code, Gemini 2.5 Pro 63.8 % using the designated agent’s preparation. The possibilities of thinking about the model also enable him to create a video game by creating an implementable symbol from one line.

Based on the strengths of its predecessors

Gemini 2.5 depends on the basic strengths of previous Gemini models, including the original multimedia and a long context window. 2.5 Pro was launched with a single symbolic context window, with plans to expand this to two million icons soon. This model enables the understanding of vast data collections and dealing with complex problems from various sources of information, texts, sound, photos, videos, and even entire programming instructions warehouses.

Developers and institutions can now start the Gemini 2.5 Pro experience in Google AI Studio. Gemini Advanced users can also access it via a model drop -down desktop and portable platforms. The model will be launched on VERTEX AI in the coming weeks.

Google DeepMind encourages users to make comments, which will be used to increase the enhancement of Gemini capabilities.

(Photography by Echta Nair)

See also: Deepseek v3-0324 leads the artificial intelligence models

Do you want to learn more about artificial intelligence and large data from industry leaders? Check AI and Big Data Expo, which is held in Amsterdam, California, and London. The comprehensive event was identified with other leading events including the smart automation conference, Blockx, the digital transformation week, and the Cyber ​​Security & Cloud.

Explore the upcoming web events and seminars with which Techforge works here.