Google has launched its latest artificial intelligence (AI) model, Gemini 2.5, optimized for solving complex problems, entering the full-scale competition for reasoning-type AI. This model significantly improved performance and accuracy by employing a 'thinking' approach prior to responses, and it has been shown to surpass the major models of competitors across several benchmarks.
Google announced on the 25th (local time) that it will release Gemini 2.5. This is an upgrade approximately three months after the release of Gemini 2.0 in December last year. Google evaluated this model as "the most intelligent model developed to date" and explained that it is designed for a reasoning structure to solve complex problems.
Gemini 2.5 goes beyond simple classification and prediction, having the ability to analyze information and draw logical conclusions. Google emphasized that accuracy and performance have improved based on advanced reasoning capabilities that can reflect context and nuance when making decisions.
The first version of this model, Gemini 2.5 Pro, is available through the Google AI Studio and a subscription service called Gemini Advanced at $20 per month. Google has also announced plans to integrate reasoning capabilities directly into all models in the future.
The benchmark performance is also noteworthy. According to Google, Gemini 2.5 Pro outperformed top models from major competitors such as OpenAI, Anthropic, and China's DeepSeek in tasks including code editing, software development, and math, science, and humanities-based multimodal tests. In particular, it was noted that it surpassed OpenAI's O3-Mini and DeepSeek's R1 models.
Additionally, this model supports a context window of 1 million tokens, capable of processing about 750,000 words at once. This exceeds the entire volume of 'The Lord of the Rings' series, and there are plans to expand this to 2 million tokens soon.
Previously, Google introduced a 'Gemini 2.0 Flash Thinking' model with integrated thinking features in January, and it is now accelerating its efforts to secure leadership in reasoning-type AI with the release of this 2.5 version.