Google Unveils Gemini 2.5 Deep Think: A Leap Ahead in AI Capabilities
In a significant announcement today, Google revealed the deployment of its Gemini 2.5 Deep Think model within the Gemini app, exclusively available to Google AI Ultra subscribers. This cutting-edge model has outperformed notable competitors, including OpenAI’s o3 and xAI’s Gork 4, across multiple essential benchmarks.
Evolution from Gemini 2.5 Pro
The journey of Gemini 2.5 began at the Google I/O 2025 event in May, where the Gemini 2.5 Pro Deep Think model was introduced. This innovative model utilized advanced research methodologies to analyze various hypotheses prior to delivering responses. Today’s update brings enhanced features, driven by valuable input from early testers and recent advancements in research. Google asserts that these updates mark a substantial leap forward compared to the initial iteration showcased during I/O.
A Bronze Medal Performance
Additionally, Google disclosed that the latest Deep Think model is a variant of the version that recently earned a gold medal at the 2025 International Mathematical Olympiad (IMO).However, to ensure broader applicability, the model has been optimized for quicker responses, albeit achieving a Bronze-level rating on the 2025 IMO benchmark.
Outstanding Benchmark Results
As highlighted in the accompanying benchmarks, Gemini 2.5 Deep Think has excelled in evaluations such as LiveCodeBench V6, Humanity’s Last Exam, IMO 2025, and AIME 2025, showcasing its robust capabilities.
Accessing Deep Think in the Gemini App
Subscribers to Google AI Ultra can now utilize Deep Think through the Gemini app, with a designated limit on daily prompts. Users can select the ‘Deep Think’ feature in the prompt bar when using the Gemini 2.5 Pro model. Furthermore, Google has indicated that Deep Think will seamlessly integrate with tools like code execution and Google Search. In the coming weeks, there are plans to introduce versions of Deep Think with and without tools to a select group of trusted developers via the Gemini API.
The Competitive Landscape
The focus now shifts to the anticipated launch of OpenAI’s GPT-5 model, which raises the pivotal question: Will it surpass the Gemini 2.5 Deep Think in forthcoming benchmark evaluations?
Leave a Reply