Google’s New Gemini 3 Deep Think Mode Is Live: Here’s What It Does And How To Use It

Google is rolling out another major upgrade inside the Gemini app, and this one is aimed at people who deal with tougher, more layered tasks. It’s called Gemini 3 Deep Think, and Ultra subscribers can start using it right away.

Think of it as a mode that takes a slower, more analytical approach instead of rushing to a quick answer.

A Mode Built For Harder Questions

Deep Think builds on what Google introduced last year with its earlier reasoning variants. The new version runs multiple lines of thinking in parallel, which helps when you’re dealing with problems that don’t have a straight path. So if you’re trying to solve a long math question, or you’re stuck on a logic puzzle that needs step by step analysis, the model tries a few routes before settling on an answer.

Google says this leads to more stable reasoning. Instead of guessing or jumping to conclusions, Deep Think works through the possibilities in a more structured way, which makes it feel a bit more thoughtful.

Benchmark Scores Show How Far It Has Come

Google shared a couple of numbers from well known reasoning benchmarks, and they give a sense of what Deep Think is trying to achieve. The model scored 41 percent on Humanity’s Last Exam without any external tools. With code execution, it reached 45.1 percent on ARC AGI 2.

These tests focus on pattern recognition, problem solving, and logical reasoning rather than memorization, so the improvements matter if you rely on AI for technical or academic work.

How You Can Use It Inside The Gemini App

If you’re an AI Ultra subscriber, you’ll spot Deep Think in the prompt bar. Switching to the mode is as simple as tapping on the Deep Think label and selecting Gemini 3 Pro in the model selector. The feature sits alongside the regular chat experience, so you can keep it off for simple tasks and switch it on when you need a deeper breakdown.

Part Of A Bigger Gemini 3 Push

Deep Think isn’t arriving alone. It’s another piece of the broader Gemini 3 upgrade Google announced earlier. The new models promise stronger multimodal understanding, which basically means they can work with text, images, videos, PDFs and screenshots with more accuracy.

The company also highlighted improvements in long context handling, planning and safety, all of which help the system follow instructions without losing track of what you’re trying to do.

Source