Google raises the bar with Gemini 3 Deep Think for complex problem-solving

Google has introduced Gemini 3 Deep Think, a new AI reasoning mode for Ultra subscribers that delivers deeper analysis, multi-step logic processing and significantly improved performance on complex scientific, mathematical and real-world problem-solving tasks.

Google has expanded its Gemini ecosystem with the release of Gemini 3 Deep Think, an enhanced reasoning mode designed to handle complex, multi-layered analytical tasks. The feature is now available to Google AI Ultra subscribers through the Gemini app. The company describes Deep Think as its most capable reasoning system so far, built to generate structured, methodical explanations across subjects like mathematics, science, engineering and logic.

The launch marks Google’s latest attempt to elevate AI from conversational tools to systems capable of supporting advanced research and problem-solving workflows.

Breakthrough performance in logic and benchmark tests

According to Google, Gemini 3 Deep Think uses a parallel reasoning framework that examines multiple hypotheses before selecting the most accurate solution. This allows the model to maintain context across lengthy problem chains — a major improvement over earlier versions.

Benchmark performance highlights show significant gains. On the “Humanity’s Last Exam” advanced reasoning test, Deep Think achieved a score of 41%, surpassing all previous AI models. When combined with code execution, it reached 45.1% on ARC-AGI-2, a benchmark considered one of the toughest indicators of machine reasoning capability. These results extend progress first demonstrated by Gemini 2.5 Deep Think, which earlier matched human-level achievements in international contests such as the International Mathematical Olympiad.

Google says the intention behind Deep Think is to move AI beyond short, chat-style responses and toward models capable of explaining their reasoning — a key requirement for scientific, analytical and professional use cases.

How users can access the new mode

Gemini 3 Deep Think is already live for those on the Ultra subscription tier. Users can access it by selecting the “Deep Think” toggle within the Gemini app and choosing Gemini 3 Pro as the active model. The feature works across devices and supports both everyday question-answering and highly technical computation.

Announcing the rollout on X, Google stated that Deep Think “explores multiple hypotheses simultaneously to deliver more sophisticated output,” signalling a deeper shift toward reasoning-driven AI systems.

With competition rising across the global AI landscape, industry analysts view this update as a strong signal of Google’s ambition to reclaim leadership in advanced reasoning technologies. By integrating Deep Think into Gemini 3, the company aims to offer researchers, developers and students an AI co-pilot capable of handling complex, multi-step challenges with greater precision than ever before.