Google launched Gemini 2.5 on Tuesday as a next-generation AI reasoning tool built to boost problem-solving performance. The new version of Google’s AI models represents a major advancement in the field by integrating sophisticated reasoning techniques that enhance both accuracy and performance.

Introducing Gemini 2.5 Pro Experimental

Google announced Gemini 2.5 Pro Experimental as the flagship model of its Gemini 2.5 family, which they assert to be the most intelligent reasoning model they have developed so far. Subscribers who pay $20 monthly for Google’s Gemini Advanced plan can access the model through Google AI Studio and the Gemini app.

The current release demonstrates Google’s dedication to AI reasoning, while future models will be required to include these functionalities by default.

The Competitive Landscape of AI Reasoning Models

The tech industry launched a competition to develop better AI reasoning models after OpenAI introduced o1 as the first model of its kind in September 2024. Anthropic, alongside DeepSeek, Google, and xAI, are competing to create powerful models that use enhanced computational capabilities to verify facts and solve complex questions before providing responses.

AI reasoning models demonstrate exceptional performance in mathematics and coding areas, which allows AI systems to address more intricate problem-solving activities. According to industry experts, reasoning models will become essential for developing AI agents because these agents can perform tasks with minimal human oversight. The advancement in reasoning capabilities requires these models to use more computational power, which results in higher operational costs.

Google’s Progress with AI Reasoning Models

Google’s latest attempt to outperform OpenAI’s “o” series through Gemini 2.5 demonstrates its most ambitious move in AI reasoning experimentation. The company launched “thinking” Gemini earlier in December, but the newest update leads to major enhancements in reasoning capabilities as well as computational efficiency.

Performance Benchmarks: How Gemini 2.5 Pro Stacks Up

Google asserts that its Gemini 2.5 Pro model exhibits superior performance compared to previous AI models they developed and many top competitors based on multiple industry benchmarks.

1. Code Editing: Aider Polyglot Evaluation

Google emphasizes the Aider Polyglot benchmark as a critical performance test for AI code editing evaluation. Google’s Gemini 2.5 Pro model achieved a score of 68.6% and surpassed the computational models developed by OpenAI, Anthropic, and DeepSeek.

2. Software Development: SWE-bench Verified Test

The SWE-bench Verified evaluation, which tests software development capabilities, showed that Gemini 2.5 Pro achieved a score of 63.8%. The model is superior to OpenAI’s o3-mini and DeepSeek’s R1 but remains behind Anthropic’s Claude 3.7 Sonnet, which leads with a score of 70.3%.

3. Multimodal Testing: Humanity’s Last Exam

The Gemini 2.5 Pro achieved 18.8% on Humanity’s Last Exam, which tests knowledge across mathematics, humanities, and natural sciences, while surpassing most flagship AI models.

Revolutionary Context Window Expansion

The main breakthrough of Gemini 2.5 Pro involves its expanded context window capacity that reaches 1 million tokens, enabling the model to handle up to 750,000 words during one session. The model’s context window holds more content than the complete “Lord of The Rings” book collection. Google intends to extend the input length soon to achieve a model context window of 2 million tokens.

Pricing and Availability

As of now, Google has launched Gemini 2.5 Pro but has not yet made public the pricing details for its API. Further details will be released by the company during the upcoming weeks.

Final Thoughts

The release of Google’s Gemini 2.5 Pro marks a significant advancement in AI development towards models that prioritize reasoning capabilities. The development of AI systems that incorporate a reflective pause before responding will establish new benchmarks for precision and trustworthiness. Google emerges as a strong rival in achieving AI dominance thanks to Gemini 2.5 Pro’s enhanced benchmarks and the significant expansion of its context window.