Google presented its new artificial intelligence innovation—Gemini 2.5 on Tuesday as a family of reasoning AI models that stop processing information before responding. This development represents a major advancement in AI research since Google strives to embed reasoning abilities into all upcoming models.

Introducing Gemini 2.5 Pro Experimental

Google is introducing Gemini 2.5 Pro Experimental to start its new AI series while presenting a multimodal artificial intelligence model with improved reasoning features. The business states that this new model represents their highest intelligence achievement so far and outperforms earlier models across multiple benchmarks.

Developers will have access to Gemini 2.5 Pro Experimental through Google AI Studio. The Gemini app will provide access to Gemini 2.5 Pro Experimental for users who have subscribed to the Google Gemini Advanced plan, which costs $20 per month. The arrangement makes sure that developers, together with AI enthusiasts, will personally experience the new capabilities.

Google’s new release puts them in direct competition against industry leaders like OpenAI, Anthropic, DeepSeek, and xAI, which have developed AI reasoning models. The models utilize enhanced computing capacity to validate information and navigate through complicated problems whenever they create responses.

AI Reasoning Models Are Leading to Industry Transformations

The introduction of AI reasoning models has brought remarkable changes to mathematical problem-solving and coding tasks. The development of these technologies has sparked conversations regarding their application in AI agents that operate independently with minimal human oversight. Reasoning-based models deliver better accuracy and functionality, but they require more computational resources.

Google introduced a reasoning-enhanced AI system named “thinking” Gemini in December 2024. Gemini 2.5 represents Google’s largest effort to outperform OpenAI’s “o” series, which was first launched with o1 in September 2024.

Performance Benchmarks: How Gemini 2.5 Compares

According to Google, Gemini 2.5 Pro has surpassed both its predecessor models and multiple competing AI systems during key performance evaluations. The optimization of this model enhances its capabilities in web app development and agentic coding, which positions it as a valuable tool for software developers.

Key performance metrics include:

Aider Polyglot Benchmark (Code Editing Ability):

The Gemini 2.5 Pro model achieved a 68.6% score, which exceeded the performance of competing models from OpenAI, Anthropic, and DeepSeek.

SWE-bench Verified (Software Development Ability):

Gemini 2.5 Pro achieved a score of 63.8% while demonstrating better performance compared to OpenAI’s o3-mini and DeepSeek’s R1. The performance of Gemini 2.5 Pro did not surpass Anthropic’s Claude 3.7 Sonnet, which earned 70.3%.

Humanity’s Last Exam (Multimodal General Knowledge Test):

Gemini 2.5 Pro achieved an 18.8% performance score that topped its competition among other flagship models.

The data demonstrates Google’s dedication to enhancing its AI models to improve logical reasoning accuracy and problem-solving functionality.

Unprecedented Token Context and Future Enhancements

The Gemini 2.5 Pro’s standout feature includes a massive 1 million token context window, which enables it to process around 750,000 words in one input. The model’s 1 million token context window exceeds the total length of the entire Lord of the Rings book series. The expanded token limit helps the model retain context while evaluating extensive and complex documents.

Google will soon expand the input length, which will result in a token capacity increase to 2 million tokens. This development promises to transform how AI systems process massive data volumes with improved efficiency.

Google has not shared the pricing details for the Gemini 2.5 Pro API but announced that they will release more information soon.

Google’s newest AI release establishes a new standard for reasoning models while strengthening its position within the competitive AI field as AI technology progresses. Gemini 2.5 Pro Experimental shows that AI reasoning will soon become standard practice while transforming artificial intelligence interaction with sophisticated challenges and judgment tasks.