Google is preparing to launch its Deep Think feature for Gemini 2.5 Pro, which was officially teased during Google I/O 2025. This new mode is designed to work much like OpenAI’s ChatGPT o3 Pro, offering extra compute power for more challenging problems. Users who deal with complex reasoning tasks, such as advanced math, coding, and multimodal workflows, stand to benefit the most, since Deep Think targets those scenarios where standard responses might fall short.
Official benchmark results indicate that Deep Think delivers a notable 15% performance boost over the regular Gemini 2.5 Pro in math-related tasks, with improvements also seen in code and multimodal benchmarks.
The UI for Gemini web is also being updated: Google is moving the prompt bar to the center of the screen, following design choices familiar to ChatGPT, possibly to make Gemini feel more like a core search or productivity hub.
Within the new tool selector, the personalization feature, which lets users search their past chat data, has shifted from its previous location and now sits alongside the new Deep Think option. Selecting Deep Think triggers a notification that processing may take several minutes, as the model works through more intensive reasoning tasks. This slower, more thorough approach echoes the "kingfall" model seen in AI Studio, which is known for high-quality outputs in coding and creative image generation.
While there’s no confirmed public release date, signs point to an imminent launch, with the feature and related announcements starting to surface in the interface. There’s also speculation that Deep Think might initially be limited to Ultra-tier users, but this remains unconfirmed. For Google, Deep Think represents another step in positioning Gemini as a technical tool for professionals who need more robust AI-powered problem-solving.
