On Tuesday, Google introduced Gemini 2.5, a new family of AI reasoning models that deliberately “pauses to think” before generating responses. In addition, the company launched Gemini 2.5 Pro Experimental—a multimodal reasoning model it claims is its most intelligent yet.
This model is now available on the Google AI Studio developer platform and within the Gemini app for subscribers to the $20-a-month Gemini Advanced plan.
Google asserts that Gemini 2.5 Pro outperforms its earlier frontier models and several competing systems on a range of benchmarks. For example, in an evaluation focused on code editing (Aider Polyglot), Gemini 2.5 Pro scored 68.6%, surpassing leading models from OpenAI, Anthropic, and DeepSeek.
Meanwhile, on another test (SWE-bench Verified) that measures software development abilities, it scored 63.8%. Although this score exceeds OpenAI’s o3-mini and DeepSeek’s R1, it falls short of Anthropic’s Claude 3.7 Sonnet, which achieved 70.3%.
On a multimodal assessment called Humanity’s Last Exam—which includes thousands of crowdsourced questions spanning mathematics, humanities, and natural sciences—Gemini 2.5 Pro scored 18.8%, outperforming most rival flagship models.
Google highlights that Gemini 2.5 Pro is equipped with a 1 million token context window. In other words, the model can process roughly 750,000 words at once, which is longer than the entire text of the “Lord of the Rings” series. Looking ahead, Google plans to double this capacity to 2 million tokens.
Although Google has not yet published API pricing for Gemini 2.5 Pro, the company promises to share more details in the coming weeks.
Get the latest tech news, telecom insights, and product launches wherever you prefer.
Add ProPakistani to Preferred Sources and see more of our stories in Google Search and Top Stories.