Google launched Gemini 2.5 on Tuesday, which represents a new series of AI models aimed at boosting reasoning abilities. Current AI technology has reached a new milestone because the latest advancement lets the system take time to “think” before answering.

Google unveils Gemini 2.5 Pro Experimental as part of its latest AI model series while touting it as its smartest model to date. The Gemini 2.5 Pro Experimental model goes live Tuesday on Google AI Studio and the Gemini app, which is accessible to subscribers of Google’s Gemini Advanced service with a $20 monthly fee.

AI Reasoning Models: The Next Big Leap in Artificial Intelligence

The tech industry entered a competitive phase to create advanced reasoning models after OpenAI unveiled its first AI reasoning model, o1, in September 2024. The field of AI reasoning models saw the entry of multiple companies when Anthropic, DeepSeek, Google, and xAI released their independent models. The models use additional computing power for factual verification and comprehensive problem analysis before giving their responses.

AI systems that utilize reasoning capabilities have shown significant advancements in solving complex mathematical equations and programming challenges. Leading professionals in artificial intelligence expect these models to become essential in creating AI agents that function independently while requiring minimal human guidance. Advanced reasoning models require more resources for operation due to their complexity.

In December, Google launched a “thinking” version of Gemini as part of its AI reasoning experimentations. The Gemini 2.5 model stands as the company’s ultimate demonstration of their commitment to challenge OpenAI’s “o” series of models.

Performance Benchmarks and Capabilities

Google’s testing shows that Gemini 2.5 Pro outperforms previous Google AI models, along with many top competition models in benchmark assessments. The company optimized this model specifically for building visually engaging web applications and agentic coding solutions.

The performance evaluation Aider Polyglot tested code editing capabilities and awarded Gemini 2.5 Pro with a score of 68.6%. The performance of Gemini 2.5 Pro exceeded that of leading AI models developed by OpenAI, Anthropic, and the Chinese research lab DeepSeek.

Gemini 2.5 Pro received a 63.8% score from SWE-bench Verified, which measures software development skills. Gemini 2.5 Pro exceeded the performance of OpenAI’s o3-mini and DeepSeek’s R1 but did not match the score of 70.3% achieved by Anthropic’s Claude 3.7 Sonnet.

Gemini 2.5 Pro achieved an 18.8% score on Humanity’s Last Exam, which included thousands of crowdsourced questions across mathematics, humanities, and natural sciences and surpassed most other flagship models.

Expanding Context Window and Future Updates

The Gemini 2.5 Pro will come with a 1 million token context window that lets the AI handle around 750,000 words at launch—a capacity exceeding the length of “Lord of The Rings.” Google intends to soon double the input capacity by expanding the context window to 2 million tokens.

The company has not released the Gemini 2.5 Pro API pricing after demonstrating its advanced features. The company announced that pricing information will be released soon.

Through this new release, Google confirms its commitment to advancing AI reasoning, which promises future AI systems will deliver improved accuracy and problem-solving abilities. Tech giants continue advancing AI frontiers as competition grows because reasoning models will form the backbone of future intelligent systems.