Google Introduces Gemini 2.5 AI Reasoning Models

On Tuesday, Google unveiled Gemini 2.5, a new family of artificial intelligence (AI) reasoning models designed to “think” before responding to questions.

Introducing Gemini 2.5, our most intelligent AI model.

Our first release, an experimental version of 2.5 Pro, unlocks state-of-the-art performance in math and science.

Learn more pic.twitter.com/aoe7egliJb

— Google (@Google) March 25, 2025

As part of this launch, the company introduced Gemini 2.5 Pro Experimental, a multimodal reasoning AI model that it describes as its most advanced yet.

The model is now accessible through Google AI Studio and the Gemini app for subscribers to Gemini Advanced, priced at $20 per month.

Pioneering Reasoning in AI Models

Google has announced that all future AI models will integrate reasoning capabilities.

This comes in the wake of OpenAI’s groundbreaking launch of the o1 model in September 2024, which set off a race among tech companies—including Anthropic, DeepSeek, Google, and xAI—to develop similarly advanced models.

AI reasoning models stand out for their ability to fact-check and analyse problems using additional computational power and time before presenting solutions.

Read Also: Google Unveils Experimental AI-Powered Search Engine Enhancement

These reasoning have contributed to new milestones in tasks like mathematics and coding.

Experts suggest that reasoning models will play a critical role in the development of AI agents capable of functioning autonomously.

However, these advancements come at a higher cost.

Benchmark Performance and Competitive Edge

While Google has experimented with reasoning in previous iterations, Gemini 2.5 represents its most robust challenge to OpenAI’s “o” series.

The company claims that Gemini 2.5 Pro surpasses previous frontier models and many competitors in several benchmarks.

For instance, on the Aider Polyglot evaluation for code editing, Gemini 2.5 Pro achieved a score of 68.6%, outperforming leading models from OpenAI, Anthropic, and DeepSeek.

On SWE-bench Verified, which evaluates software development capabilities, it scored 63.8%, surpassing OpenAI’s o3-mini and DeepSeek’s R1 but falling short of Anthropic’s Claude 3.7 Sonnet, which scored 70.3%.

In the comprehensive Humanity’s Last Exam, Gemini 2.5 Pro scored 18.8%, outperforming most competing flagship models.

Enhanced Capabilities and Token Context

The Gemini 2.5 Pro model boasts a 1 million token context window, enabling it to process approximately 750,000 words in a single session—equivalent to the length of the “Lord of the Rings” trilogy.

Google has announced plans to extend this capacity to 2 million tokens in the near future.

Pricing and Future Updates

While the company has not yet disclosed API pricing for Gemini 2.5 Pro, it has indicated that more details will be provided in the coming weeks.

This latest development underscores Google’s commitment to advancing AI reasoning capabilities and positioning itself as a leader in the competitive AI landscape.