Google has introduced a new Deep Think mode for its Gemini 2.5 Pro, enhancing its AI’s ability to consider multiple reasoning paths before responding.
This mode allows Gemini to run multiple possible lines of reasoning in parallel before deciding how to respond, resembling looking both ways or rereading instructions before building furniture.
Google’s tests have shown that Deep Think’s brainpower is working, with it performing at a top-tier level on the 2025 U.S. math olympiad and scoring an impressive 84% on the popular MMMU.
Deep Think is not yet widely available, but it will be available once all kinks are resolved. Gemini 2.5 Pro now supports native audio generation out, allowing it to talk back to users and has an “affective dialogue” feature that detects emotional shifts in tone and adjusts accordingly.
The Deep Think mode is part of Google’s growing lineup of features and APIs used by developers to embed Gemini in their software.
