Google launches newest AI model Gemini 2.5

Google of the United States announced on March 25 that it has developed the latest basic model of generative AI (artificial intelligence), “Gemini 2.5”. Compared with previous models, it has improved the logical thinking ability of deducing answers in sequence. The company will use this to compete with OpenAI of the United States and DeepSeek, a Chinese startup that focuses on logical thinking models.

“Gemini 2.5 Pro” was released on the 25th as one of the new models. It will first be provided through the platform for developers and the paid version of Gemini. Among Google’s models, “Gemini 2.5 Pro” has the strongest ability to handle complex topics such as math problems and programming.

General conversational AI will immediately answer users’ questions based on the knowledge learned in advance. However, the logical thinking model will not answer questions immediately, but will take time to logically sort out and answer after deep thinking. Google has also been developing a logical thinking model in the existing model “Gemini 2.0”.

In terms of logical thinking models, OpenAI and DeepSeek have always been in the leading position. According to Google, the performance of the “Gemini 2.5 Pro” released this time in solving mathematical problems exceeds that of OpenAI’s “o3-mini” and DeepSeek’s “R1”.