Google launched Gemini 2.5 on Tuesday, which represents a new series of AI models constructed to improve reasoning functionality. The new AI advancement permits systems to take a moment to deliberate before delivering responses, which represents a major technological breakthrough in artificial intelligence.

Google launches Gemini 2.5 Pro Experimental to showcase their most intelligent multimodal reasoning AI model yet. The model will launch on Tuesday, and subscribers to Gemini Advanced can access it via Google AI Studio and the Gemini app.

AI Reasoning Models: The Next Big Leap in Artificial Intelligence

After OpenAI released its initial AI reasoning model named o1 in September 2024, the technology sector began competing to create advanced reasoning models. Anthropic, DeepSeek, Google, and xAI have introduced their proprietary AI reasoning models to compete in the industry. These models utilize additional computing resources to check facts and conduct thorough problem analysis before generating an answer.

AI systems that utilize reasoning have shown significant advancements in tackling intricate mathematical equations as well as coding problems. The development of AI agents that function autonomously and require little human supervision is expected to heavily depend on these reasoning models, according to many experts. The advanced nature of reasoning models results in higher operating costs.

In Decembe,r Google introduced a “thinking” version of Gemini after experimenting with AI reasoning. Gemini 2.5 marks the company’s most advanced dedicated effort to rival OpenAI’s “o” series of models.

Performance Benchmarks and Capabilities

Google claims that their Gemini 2.5 Pro model outperforms both their earlier AI systems and numerous competing models in multiple benchmark evaluations. The company created this model specifically to achieve superior performance in building visually appealing web applications alongside agentic coding programs.

The Aider Polyglot evaluation that tests code-editing capabilities recorded Gemini 2.5 Pro with a 68.6% score. Gemini 2.5 Pro displayed superior performance compared to top AI models developed by OpenAI and Anthropic, along with DeepSeek’s Chinese research laboratory.

Gemini 2.5 Pro received a score of 63.8% in the SWE-bench Verified software development skills test. Gemini 2.5 Pro achieved better results than OpenAI’s o3-mini and DeepSeek’s R1 but was unable to reach Anthropic’s Claude 3.7 Sonnet score of 70.3%.

When tested with Humanity’s Last Exam, which includes thousands of crowdsourced questions across mathematics, humanities, and natural science, Gemini 2.5 Pro achieved an 18.8% score while most competing flagship models fell behind.

Expanding Context Window and Future Updates

The launch of Gemini 2.5 Pro will include a context window of 1 million tokens, enabling the processing of about 750,000 words per session, which exceeds the full text length of “The Lord of the Rings” book series. Google intends to soon expand this input capacity by doubling the context window to 2 million tokens.

Google has presented its new advanced features, but has yet to provide information about the Gemini 2.5 Pro API pricing. The organization will provide further details about pricing in the upcoming weeks.

Google’s most recent release demonstrates its commitment to advancing AI reasoning to achieve enhanced accuracy and analytical capabilities with complex problem-solving abilities. Tech giants respond to intensifying competition by advancing AI capabilities, which position reasoning models as essential elements for future intelligent systems.