- Google Gemini 2.5 Pro is getting a new deep think model
- Allows Gemini to consider numerous reasoning routes before responding deep think
- Deep Think will improve Gemini’s accuracy on complex math and code
Google is adding some extra mental strength to Gemini with a new deep think mode. The company unveiled the latest option for Google I/O 2025 for Google Gemini 2.5 Pro, showing what its AI can do with extra depth.
The deep think mainly enhances Gemini’s ‘brain’ with extra brain. In a deep think mode, Gemini will not spit the answer to a question as soon as possible. Instead, it operates various potential lines of reasoning parallel before making this decision. This is equivalent to seeing both methods, or re -reading the instructions before making a piece of furniture.
And if Google’s tests are anything, the mental strength of the Deep Think is working. It is performing at an advanced level at the 2025 -math Olympiad, which is at the top of the Leo Kodinch’s competitive programming test, which scored a surprisingly 84 % of the most surprisingly 84 % on the popular MMU, which is a kind of multi -model reasoning work. Deep thinking is not yet widely available. Google is just sending it in front of reliable testers for now. But, dominant. , Once all the concrete is ironed, everyone will have deep access to Gemini’s ideas.
Gemini shines
Deep Think Fits the rest of the Gemini 2.5 growing lineup and its software in the software used by developers to embed Gemini in the API new features for its various models.
For example, Gemini 2.5 Pro now supports local audio generation. This means that it can talk to you. The speech features an “impressive conversation”, which detects and adjusts to the emotional changes in your tone. If you feel pressure, Gemini can quit talking like a customer service agent and responds like a friend (or at least AI translates such a reaction). And it would be better to know when to talk about the new active audio feature, which filters the background noise, so Gemini only rotates when you are sure you are talking to him.
Founded with new security security measures and upcoming project Marine Computer Use features, Gemini 2.5 is trying to trust not only with your calendar or code, but also with your book statement or the entire operating system.
Another factor spreading in Gemini 2.5 is what Google calls the ‘thinking budget’. Earlier, the Gemini 2.5 flash unique, thinking budget developers lets decide how deeply the model should think before responding. This is a good way to ensure that you get a full answer without spending too much. Otherwise, deep thinking can give you only one taste of its reasoning, or give you the whole thing and make it very expensive for any follow -up.
If it is unclear what these ideas are included, Gemini will present ‘thought summary’ for the 2.5 Pro and Flash developers, a document showing the definite details of what AI is doing in the matter of applying the information through its reasoning process, so you can really see the AI ​​inside the brain.
This indicates an axis of all models that just talk faster to emphasize those who can make deep arguments, if slow. Deep thinking is part of this change deliberately, layered reasoning. It is no longer trying to predict the next word, it is implementing this logic to theories and the process of coming up with the answers to your questions is very high. Google seems to be able to enable Gemini not only to bring answers, but is capable of understanding the shape of the question itself.
Of course, the AI ​​argument is still in a place where a very logical answer can come with a random aspect of nonsense, no matter how impressive the benchmark scores are. But you can begin to see the shape of what you are coming, where the promise of a real ‘co -pilot’ results.