On March 26, 2025, Google unveiled its latest artificial intelligence model, Gemini 2.5, claiming it to be the most intelligent iteration yet, surpassing the capabilities of OpenAI's o3. This announcement marks a significant milestone in the ongoing competition within the AI landscape, as Google positions itself at the forefront of advanced AI technology.
Gemini 2.5 Pro Experimental, the flagship version of this new model, is now available for users subscribed to the Gemini Advanced plan and can be accessed via Google AI Studio. Google has made a bold statement, asserting that all future AI models will be reasoning models, a shift that reflects the company's commitment to enhancing the cognitive capabilities of its AI systems.
The introduction of reasoning capabilities in AI models is not entirely new; the first such model, known as o1, was launched in September 2024. However, Gemini 2.5 is seen as Google's most serious attempt to compete with OpenAI's advanced models, particularly in the realm of multimodal reasoning.
According to Google, Gemini 2.5 Pro has outperformed its predecessors and several leading AI models from competitors across various benchmarks. For instance, in the Aider Polyglot benchmark, which assesses coding skills, Gemini 2.5 Pro achieved a score of 68.6%, surpassing the top models from OpenAI, Anthropic, and DeepSeek. On the SWE-bench Verified test, it scored 63.8%, outperforming OpenAI's o3-mini and DeepSeek's R1, although it fell short of Anthropic's Claude 3.7 Sonnet, which scored 70.3%.
Moreover, on the comprehensive Humanity’s Last Exam, which contains a vast array of tasks spanning mathematics, humanities, and natural sciences, Gemini 2.5 Pro achieved a score of 18.8%, better than most flagship models from competitors. These results suggest that Gemini 2.5 Pro is not just a step forward; it represents a leap in performance and capability.
One of the standout features of Gemini 2.5 Pro is its context window, which currently accommodates 1 million tokens—equivalent to approximately 750,000 words. Google plans to double this capacity soon, allowing for even more extensive data processing and enhanced task management.
In comparison to its predecessor, Gemini 2.0, which featured a branding of "Flash Thinking," the new model does not carry the explicit "Thinking" label. Instead, users can enable a feature in the Gemini application called "Show Thinking Steps," allowing them to observe the logic behind the model's responses. This transparency aims to improve user trust and understanding of AI decision-making processes.
The advancements in Gemini 2.5 Pro extend beyond mere performance metrics; they also enhance the model's ability to tackle complex tasks. Google emphasizes that this new version excels in creating visually appealing web applications and writing code, addressing the growing demand for sophisticated programming capabilities in AI.
Google's confidence in Gemini 2.5 Pro is evident in its claims regarding the model's performance in mathematical and scientific tests. It has reportedly achieved top scores in the AIME 2025 mathematics test and the GPQA diamond for natural sciences, reinforcing its status as a leading AI model.
As for its availability, Gemini 2.5 Pro is currently accessible to Gemini Advanced subscribers and users of Google AI Studio. In the coming weeks, it will also be introduced on the Vertex AI platform, expanding its reach to a broader audience of developers and businesses.
In addition to its advanced features, Google has promised to announce pricing details for API access to Gemini 2.5 Pro shortly, enabling users to leverage its capabilities for large-scale industrial applications. This move is expected to attract a wide range of users, from startups to established enterprises, looking to integrate advanced AI into their operations.
Overall, the launch of Gemini 2.5 Pro signifies a pivotal moment in the AI sector, as Google continues to push the boundaries of what is possible with artificial intelligence. With its enhanced reasoning capabilities, superior performance on benchmarks, and user-friendly features, Gemini 2.5 Pro is poised to redefine how AI can be utilized across various fields.
As the competition among AI developers intensifies, Gemini 2.5 Pro not only sets a new standard for performance but also raises the bar for future innovations in the rapidly evolving world of artificial intelligence.
