Google unveiled Gemini 2.5 on Tuesday as an innovative AI reasoning model that enhances problem-solving abilities. Google’s newest AI models achieve substantial progress in their field through the integration of enhanced reasoning methods, which boost both accuracy and performance.

Introducing Gemini 2.5 Pro Experimental

Google launched the Gemini 2.5 family with the release of Gemini 2.5 Pro Experimental, which represents the company’s most intelligent multimodal AI reasoning model yet. Google’s AI Studio and Gemini app provide access to this model, which remains exclusive to those who subscribe to the $20 monthly Gemini Advanced plan.

Google shows determination toward AI reasoning by planning to include these features in upcoming models.

The Competitive Landscape of AI Reasoning Models

The tech sector began a competitive rush to create advanced reasoning AI models after OpenAI introduced its first AI reasoning model, o1, in September 2024. The AI competition sees Anthropic, DeepSeek, Google, and xAI developing models that employ stronger computing power to validate information and solve complex problems before providing answers.

AI reasoning models have demonstrated their effectiveness in fields such as mathematics and coding, which allows AI systems to address intricate problem-solving challenges. According to industry experts, reasoning models will become essential components in developing AI agents designed to operate independently with little human oversight. The enhanced reasoning ability of these models demands greater computational resources, which results in higher operational expenses.

Google’s Progress with AI Reasoning Models

Gemini 2.5 stands as Google’s most dedicated attempt yet to outperform OpenAI’s “o” series models through its AI reasoning developments. Gemini first appeared as a “thinking” version in December but now offers significant improvements in reasoning capacity and computational efficiency.

Performance Benchmarks: How Gemini 2.5 Pro Stacks Up

Google reports that Gemini 2.5 Pro surpasses both their older AI models and numerous top competitors in multiple industry-standard tests.

1. Code Editing: Aider Polyglot Evaluation

Google emphasizes the Aider Polyglot benchmark as one of its essential tests to measure the performance of AI systems in code editing activities. Gemini 2.5 Pro scored 68.6% and demonstrated superior performance compared to AI models from OpenAI, Anthropic, and DeepSeek.

2. Software Development: SWE-bench Verified Test

The SWE-bench Verified performance test for software development capabilities gave Gemini 2.5 Pro a score of 63.8%. Despite outperforming OpenAI’s o3-mini and DeepSeek’s R1, Gemini 2.5 Pro remains behind Anthropic’s Claude 3.7 Sonnet, which maintains the highest score at 70.3%.

3. Multimodal Testing: Humanity’s Last Exam

The Gemini 2.5 Pro achieved an 18.8% result on Humanity’s Last Exam, which tests multimodal knowledge in mathematics, humanities, and natural sciences, surpassing almost all current leading AI systems.

Revolutionary Context Window Expansion

The Gemini 2.5 Pro model achieved a significant breakthrough with its 1 million token context window capability, which enables it to handle up to 750,000 words during one use session. When compared to Lord of the Rings book series, this model can handle text content that surpasses the entire trilogy in length. Google plans to increase the input length soon by expanding the model’s context window to handle 2 million tokens.

Pricing and Availability

Google has released Gemini 2.5 Pro but has yet to reveal the pricing details for its API. The organization intends to release additional information during the following weeks.

Final Thoughts

The launch of Google’s Gemini 2.5 Pro signifies a key advancement in the AI sector’s transition toward models focused on reasoning capabilities. The development of AI systems that can pause to analyze before responding is expected to establish new benchmarks for accuracy and reliability. Improved benchmarks, along with a greatly expanded context window, enable Google to become a strong challenger in the AI supremacy battle through Gemini 2.5 Pro.