Google unveiled the Gemini 2.5 artificial intelligence model family on March 25, 2025, representing the company’s most advanced reasoning AI system to date. The first released version, Gemini 2.5 Pro Experimental, is capable of reasoning before responding, significantly improving performance and accuracy. The model is already available on the Google AI Studio developer platform and for Gemini Advanced subscribers through the Gemini app.
The Gemini 2.5 Pro delivers outstanding performance across various benchmark tests, particularly in programming and complex problem-solving. It achieved a 68.6% score on the Aider Polyglot test, which measures code-editing capabilities, surpassing its competitors. On the SWE-bench Verified test, which evaluates software development skills, it scored 63.8%, outperforming OpenAI’s o3-mini (49.3%) and China’s DeepSeek R1 (49.2%), though falling short of Anthropic’s Claude 3.7 Sonnet, which scored 70.3%. In the Humanity's Last Exam test, assessing scientific and reasoning abilities, it achieved an 18.8% score, also a leading result. According to Koray Kavukcuoglu, Technical Director of Google DeepMind, the Gemini 2.5 models are capable of reasoning through their thoughts before responding, resulting in enhanced performance and improved accuracy.
The Gemini 2.5 Pro comes with a 1 million token context window, enabling the simultaneous processing of approximately 750,000 words—longer than the entire Lord of the Rings book series—with Google announcing that this will soon expand to 2 million tokens. The model excels particularly in creating web applications and autonomous code generation, capable of producing executable video game code from a single-line instruction. The technology features native multimodality, allowing it to process text, audio, images, video, and entire code repositories. According to Google, these reasoning capabilities will become a default feature in all their future AI models to support more complex, context-responsive, and autonomous AI systems. While specific API pricing has not yet been disclosed, the company has promised to release the pricing structure in the coming weeks, which will include higher rate limits to accommodate large-scale usage.
Sources:
1.
Maxwell Zeff
2.
Koray Kavukcuoglu
3.
