Google launched Gemini 2.5 on Tuesday as an advanced family of AI models to boost reasoning performance. The AI technology now includes the ability to take thinking pauses before responding, which represents a major advancement within artificial intelligence.

Google is unveiling its latest AI creation, Gemini 2.5 Pro Experimental, as part of a new generation of models with claims that it represents their most intelligent system so far. Starting Tuesday, the Gemini 2.5 Pro Experimental model will be accessible both via Google AI Studio and the Gemini app for users with a subscription to Gemini Advanced, which costs $20 per month.

AI Reasoning Models: The Next Big Leap in Artificial Intelligence

The tech industry has launched a competitive development of new models with improved reasoning abilities since OpenAI introduced their first AI reasoning model, o1, in September 2024. Multiple companies, including Anthropic, DeepSeek, Google, and xAA, have developed their own AI reasoning models to enter this field. These models utilize additional computational resources to perform detailed fact-checking and comprehensive problem analysis before delivering a response.

AI models that employ reasoning capabilities exhibit significant advancements in tackling complex mathematical equations and programming problems. A significant number of experts think these models will become essential in creating AI agents that function autonomously with limited human oversight. Advanced reasoning models demand higher operational costs due to their complexity.

Google released a version of Gemini that demonstrated AI reasoning capabilities back in December. Gemini 2.5 demonstrates Google’s most sophisticated and targeted initiative to challenge OpenAI’s “o” series models.

Performance Benchmarks and Capabilities

Google tested Gemini 2.5 Pro against previous AI models from their portfolio as well as top competitor systems and found it achieved superior results in multiple benchmark evaluations. The model was built to achieve top results in creating visually rich web applications and autonomous coding programs.

During the Aider Polyglot evaluation that tests code editing performance, Gemini 2.5 Pro received a score of 68.6%. Gemini 2.5 Pro achieved better results than top AI models from OpenAI, Anthropic, alongside the Chinese lab DeepSeek.

Gemini 2.5 Pro received a score of 63.8% in a separate software development skills test using SWE-bench Verified. The performance of Gemini 2.5 Pro exceeded that of OpenAI’s o3-mini and DeepSeek’s R1 but did not match Anthropic’s Claude 3.7 Sonnet, which achieved a score of 70.3%.

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

Expanding Context Window and Future Updates

The AI system Gemini 2.5 Pro will begin operations with a context window that processes 1 million tokens, which equals around 750,000 words during one session, surpassing the length of the complete “Lord of the Rings” book series. Google declared its intention to expand input capacity by doubling the context window to 2 million tokens soon.

Google announced the advanced features of Gemini 2.5 Pro but withheld information regarding its API pricing. The company announced that pricing details will be made available over the next few weeks.

Google’s new release represents a major investment in AI reasoning technology, which promises to make artificial intelligence systems more precise and effective at solving complex problems. Intensifying competition among tech giants leads to advancements in AI boundaries, with reasoning models becoming essential for future intelligent systems development.