Google revealed Gemini 2.5 as a new AI reasoning model on Tuesday that delivers enhanced problem-solving abilities. The new version of Google’s AI models represents significant progress within the field because it combines advanced reasoning techniques that enhance both accuracy and performance.

Introducing Gemini 2.5 Pro Experimental

Google initiated its Gemini 2.5 family by releasing Gemini 2.5 Pro Experimental, a multimodal AI reasoning model that Google states represents its highest intelligence level yet. Subscribers of Google’s $20-per-month Gemini Advanced plan have exclusive access to the model through Google AI Studio and the Gemini app.

Google demonstrates its dedication to AI reasoning through this release, while future models will incorporate these capabilities by default.

The Competitive Landscape of AI Reasoning Models

The tech industry began competing to build better reasoning AI models following OpenAI’s introduction of o1 as the first AI reasoning model in September 2024. Anthropic, DeepSeek, Google, and xAI participate in the AI reasoning competition by developing models that employ enhanced computing power to verify facts and process complex problems before providing responses.

AI reasoning models have demonstrated a high level of effectiveness in the fields of mathematics and coding, which permits AI systems to address more advanced problem-solving challenges. Experts within the industry predict that reasoning models will be essential for creating AI agents that operate autonomously to achieve tasks with little human oversight. As these models achieve better problem-solving abilities, they become more expensive to run because of their substantial computational requirements.

Google’s Progress with AI Reasoning Models

Google developed Gemini 2.5 as its most ambitious project yet to outpace OpenAI’s “o” series models, building upon previous AI reasoning experiments. The organization initially launched Gemini’s “thinking” version in December before launching an update that features substantial improvements in reasoning abilities and computational efficiency.

Performance Benchmarks: How Gemini 2.5 Pro Stacks Up

Google’s Gemini 2.5 Pro demonstrates superior performance compared to its previous AI models as well as various leading competitors across industry-standard benchmarks.

1. Code Editing: Aider Polyglot Evaluation

Google points to the Aider Polyglot benchmark as one of its essential tests for evaluating AI performance in code editing. The Gemini 2.5 Pro reached a performance level of 68.6%, surpassing AI models from OpenAI, Anthropic, and DeepSeek.

2. Software Development: SWE-bench Verified Test

Gemini 2.5 Pro achieved a 63.8% score in the SWE-bench Verified evaluation that tests AI software development skills. The Gemini 2.5 Pro surpasses OpenAI’s o3-mini and DeepSeek’s R1 but remains below Anthropic’s Claude 3.7 Sonne,t which holds the highest score of 70.3%.

3. Multimodal Testing: Humanity’s Last Exam

The rigorous multimodal test Humanity’s Last Exam assessed AI proficiency in mathematics, humanities, and natural sciences, where Gemini 2.5 Pro achieved 18.8% and surpassed most leading AI models in the market.

Revolutionary Context Window Expansion

Gemini 2.5 Pro features a revolutionary 1 million token context window that lets it handle up to 750,000 words during one continuous interaction. The model can process up to 750,000 words which exceeds the total length of “Lord of the Rings” book series. Google intends to expand the model’s context window to 2 million tokens by doubling its current input length.

Pricing and Availability

Google currently offers Gemini 2.5 Pro but remains silent about its API pricing structure. The company confirmed that additional details will be released within the next few weeks.

Final Thoughts

The release of Gemini 2.5 Pro marks a significant advance in the AI sector’s progression toward rationality-focused models. The evolution of AI will reach new standards of accuracy and reliability once systems can pause to “think” before responding. Improved benchmarks and an expanded context window enable Gemini 2.5 Pro to establish Google as a major contender in the AI supremacy race.