Google reveals a model of artificial intelligence thinking from the next generation

On Tuesday, Google unveiled Gemini 2.5, a new family of thinking models that stop “thinking” before answering a question.

To start the new models family, Google launches Gemini 2.5 Pro experimental, a multimedia Amnesty International model, which the company claims to be its most intelligent model so far. This model will be available on Tuesday in the company’s developer, Google Ai Studio, as well as in the Gemini app for subscribers in the AI ​​plan with a $ 20 a month, Gemini Advanced.

To go forward, Google says all new artificial intelligence models will get thinking possibilities.

Since I launched Openai The first thinking model in September 2024O1, the technology industry races to match or overcome the capabilities of this model with their own. Today, Antarbur, Deepseek, Google and Xai have thinking models of artificial intelligence, which use additional computing power and time to verify facts and mind through problems before providing an answer.

Thinking techniques helped artificial intelligence models achieve new heights in mathematics and coding tasks. Many in the world of technology believe that thinking models will be a major component of artificial intelligence factors, and independent systems that can largely perform tasks human intervention. However, these models are also more expensive.

Google claims that Gemini 2.5 Pro outperforms the previous border models, and some competing artificial intelligence models, on several criteria. Specifically, Google says it designed GEINI 2.5 to outperform the creation of visually convincing web applications and agent coding applications.

In an evaluation of the evaluation code, it is called Aider Polyglot, Google says that Gemini 2.5 Pro records 68.6 %, outperforms the best artificial intelligence models from Openaii, Anthropic, and Ai Lab Labesek.

However, in another test to measure the DeV program, Swe-Benced, GIMINI 2.5 Pro features 63.8 %, outperform Openai’s O3-MINI and Deepseek’s R1, but the weak performance Claude 3.7 Sonnet, which recorded 70.3 %.

In the last humanity exam, a multimedia test consisting of thousands of collective questions related to mathematics, humanities and natural sciences, Google says that Gemini 2.5 Pro records 18.8 %, better performance than most of the leading pioneering models.

To start, Google says that Gemini 2.5 Pro is shipped using a symbolic context window of one million million, which means that the artificial intelligence model can take approximately 750,000 words in one claim. This is longer than the entire “Lord of the Rings” series. Soon, Gueini 2.5 Pro will support the length of the claim (2 million icons).

I have previously tried Google with male thinking models – which had previously launched a “thinking” from Gemini in December – but Gemini 2.5 represents the company’s most competitor in Openai O Models so far.

Google’s API pricing for Gemini 2.5 Pro.