Google, Gemini 2.5 Deep Think He opened the new generation of artificial intelligence model in the AI Ultra subscription package and opened it to selected users. With this development, Deep Think, which is Google’s most advanced logical reasoning model, can now be officially used.
The model was first introduced at the Google I/O event held in May 2025. Following a limited test process, it can now be accessed through the AI Ultra plan for 9,000 TL per month.
Google says Deep Think has advanced skills such as mathematical reasoning, versatile data processing and parallel thinking. The model can process text, visual and sound data together with the capacity to evaluate multiple entries at the same time. This increases the potential of use, especially in areas such as coding and scientific analysis.
Deep Think Model has a high level of success in mathematical reasoning
According to the company, Deep Think recently gave results at the “bronze medal” level in the tests performed at the international mathematics olympics level. In these tests, the model was able to produce fast and stable results. The “Golden Medal” level is still tested limited to researchers.
Deep Think’s performance is not only limited to theoretical mathematics. Google has also tried in advanced benchmark tests such as Livecodebench and Humanity’s Last Exam. In the comparisons, OpenAI’s O3, XAI’s Grok 4 and Gemini 2.5 Pro models were obtained on the results.
In these tests, especially in criteria such as logical consistency, speed of solution and multi -faceted assessment, Deep Think came to the fore. Google’s technical documents, the basis of this success, “parallel thinking” approach is emphasized. Thanks to this method, the model can produce results by evaluating multiple solutions at the same time.
Gemini 2.5 Deep Think’s remarkable feature is that it can work in a multimodal structure. Users can process both text and visual and voice data in a single question. This structure means a new level of use for developers and researchers working in different disciplines.
AI Ultra subscribers can access this model via Gemini mobile application or web interface. Users involved in the system can use Deep Think in a limited number of queries. Google has not announced the integer of this daily border, but is expected to be updated in the coming days.
The Deep Think, which has only been opened to access to US users, is expected to be available in more countries in the coming period. Google did not give a clear calendar on this issue, but the system’s localization and infrastructure arrangements continue.
Google predicts that this model will be an effective tool in advanced research, mathematical analysis processes and algorithmic development studies. Nevertheless, the fact that it is limited to a paid plan makes it limiting for some users.
The version of the model at the “Golden Medal” level is still shared with academic circles. This variant requires higher calculation power and is therefore provided only to certain universities and research institutions. Google has not given a definite date for the extensive use of this version.