Google Releases Gemini 2.5, Its Most Intelligent AI Model So Far

Google has unveiled Gemini 2.5, its latest AI model designed to handle complex reasoning and coding tasks. This release includes the Gemini 2.5 Pro Experimental, which has secured the top spot on the LMArena leaderboard and excels in various coding, math, and science benchmarks.

"Gemini 2.5 models are thinking models that reason through their responses, leading to improved performance and greater accuracy," explained Koray Kavukcuoglu, CTO of Google DeepMind.

Google highlights that the model’s reasoning abilities go beyond mere classification and prediction, enabling it to analyze information, make logical conclusions, and incorporate context and nuance.

The Gemini 2.5 "thinking model" outperforms other leading models, including OpenAI's o3 mini, GPT-4.5, DeepSeek-R1, Grok 3, and Claude 3.7 Sonnet, across multiple benchmarks. It also achieves a record 18.8% on Humanity’s Last Exam, a dataset created by hundreds of experts to challenge the limits of human knowledge and reasoning.

"Gemini 2.5 Pro Experimental is our most advanced model for complex tasks. It tops the LMArena leaderboard — which measures human preferences — by a significant margin, indicating a highly capable model equipped with high-quality style. 2.5 Pro also shows strong reasoning and code capabilities, leading on common coding, math and science benchmarks," Google said in a blog post.

The model is currently accessible to advanced users via Google AI Studio and the Gemini app, with plans to roll out on Vertex AI soon. Google also intends to introduce pricing for higher-rate production use in the coming weeks.

Developers and enterprises can now start using Gemini 2.5 Pro in Google AI Studio. "Looking ahead, we’re embedding these thinking capabilities into all of our models, enabling them to solve more complex problems and support even more capable, context-aware agents," Google added.

Gemini 2.5 Pro excels at building visually engaging web apps, creating agent-based code applications, and facilitating code transformation and editing. On SWE-Bench Verified, Gemini 2.5 Pro scores 63.8% with a custom agent setup.

Google also emphasized the advancements in Gemini 2.5’s context-handling. "Gemini 2.5 Pro ships today with a 1 million token context window (2 million coming soon), delivering strong performance improvements over previous generations," the company said. The model is capable of processing text, audio, images, video, and full code repositories.

Gemini 2.5 follows the recent launch of Google Gemma 3, the latest version in the Gemma family of open-weight models, which succeeds Gemma 2, released last year.

Additionally, Google recently introduced Gemini 2.0 Flash, which brings native image generation into the Gemini family. It integrates multimodal input, advanced reasoning, and natural language processing (NLP) to generate high-quality visuals.