Gemini 2.5 Deep Think Model Gemini 2.5 Deep Think

Google Dipmind officially a new model of the Jina family called “Gemini 2.5 Deep ThinkIt released. Of this model as “The most advanced Google’s reasonable modelIt is said to be able to review several ideas at the same time and choose the best answer by comparing them.

The model was first introduced at the Google I/O 2025 event and is now available as the first Google Multi-Angest Multi-Angest model. This system analyzes the questions in parallel by creating several artificial intelligence agencies. Although this research method requires much more computational resources, it usually causes the user to get more accurate answers.

Google has announced that the model has made significant progress over the version introduced in I/O and is using new reinforcement techniques to optimize reasoning. According to the technology giant, this model can solve issues that need to be needed Creativity, Planning Strategic And Progress Step by step Be helpful.

The cost of using the most advanced Google Model

Despite the very good performance, Cost of use Of these models are far greater than traditional models; That’s why companies like Google and XAI have decided to provide users only through their most expensive subscriptions.

Design users $ 250 shared Google One Ultra can access this new model through the Jimana app today.

Jamina 2.5 Deep Think performance in benchmarks

The 2.5 Deep Think Model in the Humanity’s Last Exam (Hle), which includes thousands of questions in the fields of mathematics, humanities and basic sciences, performs better than competitors and without the use of points. 34.8 % Has earned. For comparison, the Grok 4 of the XAI scored 25.4 percent and the O3 model of OpenAI scored 20.3 %.

In the field of programming, the model also had a better performance in the LiveCodebench6 test with 87.6 % score than GROK 4 (79 %) and O3 (with 72 %).

The Gemini 2.5 Deep Think automatically uses tools such as executing Google’s search engine and search engine and is able to provide much longer and more accurate answers than traditional models. Google says the model is superior to competitors in the production of web development projects both visually and details and can speed up the discovery process.

Google also plans to give a group of mathematicians and researchers to a group of mathematicians and researchers at the World Mathematical Olympiad (IMO) this year.

In the coming weeks, Google will provide this model with API to a group of developers and companies to explore its commercial and research applications.