Google Unveils Gemini 2.5 Pro: The Thinking AI Model Redefining Complexity and Accuracy

Google Announces Gemini 2.5 Pro - The Most Intelligent AI Model with Thinking Capabilities

On March 25, 2025, Google unveiled Gemini 2.5 Pro (experimental) for Advanced subscribers and developers. This model is part of a series of mid-year updates and introduces significant enhancements in reasoning and thinking capabilities.

Key Features:

• Thinking Models: All models in the Gemini 2.5 family are designed to reason through their thoughts before responding, leading to better performance and accuracy. Users can opt to 'Show thinking' to see the AI's thought process.
• Enhanced Performance: The new model combines a significantly improved base model with advanced post-training techniques, making it highly effective in complex tasks.
• Top Benchmarks: Gemini 2.5 Pro leads on various benchmarks, including LMArena for human preferences, AIME 2025 for math, and GPQA diamond for science, without using costly techniques like majority voting.
• Advanced Coding: The model excels in creating visually compelling web apps and agentic code applications, scoring 63.8% on SWE-Bench Verified, an industry standard.
• Multimodality and Context Window: It supports native multimodal capabilities (text, audio, images, video, code) and has a context window of 1 million tokens, with plans to expand to 2 million soon.

Availability:

• Gemini Advanced and Google AI Studio are the first platforms to receive Gemini 2.5 Pro, followed by Vertex AI in the coming weeks.
• Pricing and Production Use: Pricing details will be introduced soon, allowing for higher rate limits suitable for scaled production use.

User Experience:

• In the Gemini app, it replaces the previous 2.0 Pro (experimental) model and supports integration with various apps (@Gmail, @YouTube, etc.) and file uploads.