Google's New AI Can Think for 10 Minutes Straight Before Answering and It's Terrifyingly Smart

Google just dropped something that might actually change how we think about artificial intelligence. It is called Gemini 3 Deep Think, and it works completely differently from any chatbot you have used before.

Instead of instantly spitting out an answer like most AI tools, Deep Think actually pauses and thinks. We are talking minutes of processing time where the AI is working through a problem step by step, almost like a human expert sitting down with a pencil and paper.

This video from Matt Wolfe about Anthropic and Google's latest updates blew up on YouTube this week with over 90,000 views, and a huge chunk of the discussion was about how impressive Deep Think is in early testing.

The AI is currently available to Google's Ultra subscribers and is being rolled out to researchers and engineers through early API access. Google is not positioning this as a casual chatbot. This is designed for serious work: scientific research, complex engineering problems, and the kind of questions that would normally require a PhD to answer properly.

What makes Deep Think different is something called "extended reasoning." Most AI models generate answers token by token in a straight line. Deep Think can branch out, consider multiple approaches, check its own work, and even backtrack if it realizes it went down the wrong path.

Early testers are reporting that it can solve graduate-level physics problems, debug complex code that other AIs get wrong, and write mathematical proofs that actually hold up to scrutiny.

Google also launched new music and multimodal AI features alongside Deep Think, showing the company is going wide and deep at the same time. The AI race just got a lot more interesting.

As reported by Labla.org.

Source: Matt Wolfe / YouTube