Google introduces Gemini 2.5 Pro and says its their “most intelligent AI model” yet
Table of Contents
After releasing Gemma 3, a new open AI model that ranks as the second-best and delivers exceptional performance on a single GPU or TPU, Google has now announced an update to its flagship series of Gemini artificial intelligence models with the introduction of an experimental version, Gemini 2.5 Pro. According to Google, Gemini 2.5 Pro is its most intelligent model yet, featuring built-in reasoning or “thinking” capabilities.
This reasoning capability was first introduced in the Gemini 2.0 Flash Thinking Experimental AI model, unveiled in December. However, compared to 2.0 Flash Thinking, Google is no longer explicitly using the “Thinking” label. Instead, users can enable the “Show thinking” option in the Gemini app to view the model’s thought process. Gemini 2.5 Pro (experimental) is initially rolling out to Gemini Advanced and Google AI Studio, with Vertex AI set to follow in the coming weeks.
Gemini 2.5 Pro experimental leads AI benchmarks
All models in the Gemini 2.5 family combine “a significantly enhanced base model with improved post-training” for better overall performance. This is evident in common AI benchmarks, where the first model in this family, Gemini 2.5 Pro, outperforms competitors from OpenAI, Anthropic, xAI, and DeepSeek. Jeff Dean, Google DeepMind Chief Scientist, and Gemini Lead, says that Gemini 2.5 Pro has achieved the highest score of 1443 on the LMArena ranking, based on human preferences, and holds the top spot.
Additionally, on Humanity’s Last Exam, a dataset created by hundreds of subject matter experts to test the limits of human knowledge and reasoning, Gemini 2.5 Pro scored 18.8%. This makes it 4.7% higher than OpenAI’s o3-mini, which scored 14% on the same test, and about 6% higher than DeepSeek R1’s 8.6% score.
Gemini 2.5 is a major leap over 2.0
In addition to highlighting Gemini’s native multimodality, which allows it to interpret not just text but also audio, still images, video, and code, Gemini 2.5 Pro features a 1 million token context window, with a 2 million token limit “coming soon” to handle larger amounts of data. Google says it’s jumped forward in quality thanks to enhanced reasoning capabilities that process tasks step by step, leading to more informed decisions and improved responses for complex prompts.
With the launch of Gemini 2.5, it is clear that we are entering what Nvidia CEO Jensen Huang calls “the beginning of the reasoning AI era.” While these new AI models require 100 times more computing power than older ones, they are also significantly more advanced. One demo video showcases Gemini 2.5 Pro using its reasoning capabilities to program a video game from a single prompt.
About the Author