Google launched its new AI reasoning model, Gemini 2.5, on Tuesday, which boosts problem-solving abilities through a “thinking” pause before response generation. Google demonstrates substantial progress in AI development by embedding advanced reasoning abilities into its models, which leads to better accuracy and improved understanding of context.

Google initiated the unveiling of its new AI model family by releasing Gemini 2.5 Pro Experimental. The company claims that this multimodal AI model represents their smartest and most effective system yet. Developers and advanced users will have access to Gemini 2.5 Pro Experimental through Google AI Studio starting Tuesday. Subscribers of the company’s $20-per-month AI plan Gemini Advanced, can access it through the Gemini app.

AI Reasoning Models Are Reshaping the Industry

OpenAI’s launch of its first reasoning model, o1 in September 2024 triggered a tech industry focus on AI reasoning models. Leading AI firms, including Anthropic, DeepSeek, xAI, and most recently Google have entered the race to produce comparable models following this development. These AI systems that focus on reasoning need more computing power and longer processing times to validate information and improve their problem-solving capabilities prior to giving an answer.

AI has advanced significantly in complex fields like mathematics and coding following the application of reasoning techniques. Tech community experts strongly believe that reasoning models will become essential for developing autonomous AI agents that operate with minimal human oversight. These models demand more computational power, which leads to higher operating costs.

Google’s Efforts to Improve AI Reasoning

Google has previously ventured into the field of reasoning-based artificial intelligence. The company unveiled an experimental Gemini version with basic “thinking” features in December 2024. Gemini 2.5 stands as Google’s most advanced effort yet to outperform OpenAI’s well-respected “o” series models.

Google states that its Gemini 2.5 Pro model surpasses previous AI models from their cutting-edge lineup and multiple leading competitor models in all tested benchmarks. The company tailored Gemini 2.5 for outstanding performance in generating visually appealing web applications and coding-intensive agentic applications.

Benchmark Performance: How Gemini 2.5 Pro Compares

Gemini 2.5 Pro received a 68.6% score on the Aider Polyglot code-editing proficiency benchmark. Gemini 2.5 Pro outperformed leading AI models developed by OpenAI and Anthropic as well as DeepSeek from China.

Testing Gemini 2.5 Pro on the SWE-bench Verified test designed for general software development proficiency, yielded a score of 63.8%. The results show Gemini 2.5 Pro performs better than OpenAI’s o3-mini and DeepSeek’s R1, but does not reach the score of Anthropic’s Claude 3.7 Sonnet, which achieved 70.3%.

Gemini 2.5 Pro achieved an 18.8% score in Humanity’s Last Exam, which comprises a multimodal test with thousands of crowdsourced questions spanning mathematics, humanities and natural sciences. Gemini 2.5 Pro leads most flagship models currently available in its category.

Expanding Context Window for Enhanced Processing

The Gemini 2.5 Pro stands out because of its wider context window. The system has the capability to handle single inputs of up to 1 million tokens, which is approximately equal to 750,000 words. If you compare it to literature, this processing capacity extends even beyond the complete “Lord of the Rings” book series. Google plans to increase their existing capacity with future support for 2 million token input lengths.

What’s Next for Gemini 2.5 Pro?

Google has yet to reveal the pricing details for Gemini 2.5 Pro’s API. The company announced plans to share additional information about pricing and commercial availability within the next few weeks.

Google demonstrated its dedication to AI reasoning technology through the launch of Gemini 2.5. The company’s new advancements demonstrate future AI models that will become more intelligent and sophisticated to solve complex problems more accurately in response to growing competition.