On Tuesday, Google made headlines by launching Gemini 2.5, a cutting-edge family of AI reasoning models that pauses to 'think' prior to providing answers, revolutionizing the way AI interacts with users.

At the forefront of this new family is Gemini 2.5 Pro Experimental, which Google touts as its most intelligent model to date. This innovative, multimodal reasoning AI model is now available on Google AI Studio and within the Gemini app for subscribers enrolled in the premium $20-a-month AI plan, Gemini Advanced.

In a bold move, Google has declared that all future AI models will be designed with inherent reasoning capabilities, marking a significant shift in the development of AI technology. The arrival of Gemini 2.5 Pro comes in the wake of OpenAI's launch of the first AI reasoning model, o1, in September 2024. Since then, a fierce race has ensued among tech giants, including Anthropic, DeepSeek, and xAI, all striving to exceed the capabilities established by OpenAI.

These AI reasoning models leverage increased computing power and time, enabling them to fact-check and logically analyze problems before arriving at a solution. As a result, they are achieving remarkable advancements in areas like mathematics and programming. Industry experts believe that reasoning AI will play a crucial role in the development of autonomous agents, systems capable of handling tasks with minimal human oversight, though these models also come with a higher price tag.

Google has previously dabbled in the AI reasoning space, with an earlier version of Gemini showcasing similar 'thinking' capabilities back in December. However, Gemini 2.5 represents a more serious and strategic challenge to OpenAI's competitive models.

Claims made by Google assert that Gemini 2.5 Pro has outperformed its prior AI models and some leading competitors across various benchmarks. The model is designed to excel in creating visually engaging web applications and sophisticated coding applications. In a recent evaluation known as Aider Polyglot, which assesses code editing proficiency, Gemini 2.5 Pro achieved a remarkable score of 68.6%, surpassing offerings from OpenAI, Anthropic, and the Chinese AI lab DeepSeek.

Conversely, in another benchmarking exercise evaluating software development skills, SWE-bench Verified, Gemini 2.5 Pro scored 63.8%. This result put it ahead of OpenAI’s o3-mini and DeepSeek’s R1, but slightly behind Anthropic's Claude 3.7 Sonnet, which topped the charts with a score of 70.3%.

In a comprehensive test called Humanity’s Last Exam, which comprised thousands of crowdsourced questions covering subjects such as mathematics and the natural sciences, Gemini 2.5 Pro achieved a score of 18.8%, outperforming many rival flagship models.

To accommodate extensive input, Google has equipped Gemini 2.5 Pro with a robust 1 million token context window—allowing it to process around 750,000 words in one instance. For context, that length exceeds the entirety of the 'Lord of The Rings' trilogy! What's more, Google plans to enhance this feature, soon supporting an impressive input length of up to 2 million tokens.

As of now, Google has not disclosed the pricing for the Gemini 2.5 Pro API. However, the tech giant promises to provide further details in the coming weeks, as anticipation builds around this state-of-the-art advancement in AI technology. Stay tuned for updates—you won’t want to miss what’s next!