CEO's Column
Search
More
Foundation Models

Google Adds ‘Deep Think’ to Gemini 2.5 Pro, Boosting AI Reasoning Power

ByRishabh Srihari
2025-05-21.about 2 months ago
Google Adds ‘Deep Think’ to Gemini 2.5 Pro, Boosting AI Reasoning Power
Google unveils “Deep Think” for Gemini 2.5 Pro at I/O 2025, enhancing AI reasoning and outperforming rivals in coding and logic benchmarks.

Google’s most advanced AI just got smarter. At I/O 2025, the company introduced Deep Think, a new enhanced reasoning mode for its flagship Gemini 2.5 Pro model.

Deep Think brings a major upgrade to Gemini’s reasoning abilities. It enables the model to explore multiple possible answers before producing a response. This approach significantly improves performance on complex benchmarks and is designed to mimic deeper human-style reasoning.

Rethinking AI Reasoning

According to Google, Deep Think pushes Gemini’s capabilities “to the limit.” While technical details remain under wraps, the system appears to use parallel reasoning techniques—likely drawing from the same research trends that power top-tier models from rivals like OpenAI.

In real-world terms, that means Gemini 2.5 Pro can now code more effectively, solve logic-driven tasks faster, and analyze more abstract queries. Google says that with Deep Think enabled, Gemini 2.5 Pro topped LiveCodeBench, a rigorous coding benchmark. It also outperformed OpenAI’s o3 model on MMMU, a test designed to measure perception and reasoning.

Also read: Google Adds ‘Deep Think’ to Gemini 2.5 Pro, Boosting AI Reasoning Power

Carefully Controlled Rollout

For now, Deep Think isn’t rolling out to everyone. It’s being made available to trusted testers through the Gemini API. Google says broader availability will follow, but only after additional safety evaluations are complete.

This cautious approach isn’t new. With growing concerns over model misuse, companies like Google are placing greater emphasis on responsible deployment of cutting-edge AI features.

Flash Gets a Speed Boost Too

Not to be left behind, Google also refreshed Gemini 2.5 Flash, its lighter, more affordable model. The update improves performance across tasks like multimodal processing, coding, and long-context comprehension. It’s also more efficient than before, making it a strong choice for developers who need speed without sacrificing too much power.

The updated Flash model is now available for preview in AI Studio, Vertex AI, and Gemini apps.

With Deep Think, Google is clearly positioning Gemini 2.5 Pro as a more competitive and capable model in the rapidly evolving AI landscape. By enhancing reasoning without compromising safety, it may just close the gap on its most advanced rivals.

Related Topics

LLMsLarge Language Models (LLMs)

Subscribe to NG.ai News for real-time AI insights, personalized updates, and expert analysis—delivered straight to your inbox.