Advertising
Layoffs in Adland: Omnicom's acquisition of IPG nears finish line. But at what human cost?
Google ups its AI game with 'Deep Think' reasoning feature
Google also emphasized that safety remains a top priority. While Deep Think has shown improved content safety compared to its predecessor, it is also more cautious and may occasionally refuse benign prompts.
Deep Think is an extension of the same technology that powered Google's Gemini 2.5 model to a gold medal at the 2025 International Mathematical Olympiad. While the competition-winning model took hours to solve intricate math problems, Deep Think is optimized for faster, everyday use, delivering what Google describes as "Bronze-level IMO performance."
Google is upping the ante in the AI race with the release of Deep Think, a new advanced reasoning feature for its Gemini app. Available to Google AI Ultra subscribers, Deep Think is designed to handle complex problem-solving by using advanced parallel thinking, promising more strategic and creative responses from the AI.
Deep Think is an extension of the same technology that powered Google's Gemini 2.5 model to a gold medal at the 2025 International Mathematical Olympiad. While the competition-winning model took hours to solve intricate math problems, Deep Think is optimized for faster, everyday use, delivering what Google describes as "Bronze-level IMO performance."
At its core, Deep Think extends Gemini's "thinking time." This allows the AI to explore multiple hypotheses simultaneously and iteratively refine its answers. Google says this approach will enable the model to tackle intricate tasks like complex coding, scientific reasoning, and algorithmic development with greater depth and accuracy.
Google believes Deep Think will be a game-changer for developers and researchers. The company has already been sharing a high-performance variant with academics to test its potential for formulating mathematical conjectures and navigating complex scientific literature. The technology also showcases major advances in reinforcement learning, which helps the AI with iterative workflows like web design and code optimization.
Google says the new feature has outperformed other models on tough benchmarks like LiveCodeBench V6 and Humanity’s Last Exam, handling coding and reasoning tasks without needing external tools.
Currently, Deep Think is available to Google AI Ultra subscribers, with a daily limit on the number of prompts. It integrates with existing Gemini features like code execution and Google Search. The company plans to offer API access to trusted developers soon.
Google also emphasized that safety remains a top priority. While Deep Think has shown improved content safety compared to its predecessor, it is also more cautious and may occasionally refuse benign prompts. The company is conducting "frontier safety evaluations" to address the risks of more advanced AI reasoning.
SPOTLIGHT
According to LinkedIn’s research with over 1,700 B2B tech buyers, video storytelling has emerged as the most trusted, engaging, and effective format for B2B marketers. But what’s driving this shift towards video in B2B? (Image Source: Unsplash)
Read MoreExplained: Standing Committee’s draft report on India’s fight against Fake News
India’s parliamentary panel warns fake news threatens democracy, markets and media credibility, urging stronger regulation, fact-checking, AI oversight and global cooperation.