rewrite this title Google Introduces New “Deep Think” Feature for Gemini 2.5 Pro Model

Google has unveiled a new thinking mode called Deep Think, added to its Gemini 2.5 Pro model. The company announced that this newly added feature achieved significant success in testing. The Deep Think feature is currently accessible via the Gemini API, though it is limited to select users for now.

Thanks to this innovation, the AI can now evaluate multiple possibilities before answering a question, leading to more accurate results. According to Google, Gemini 2.5 Pro with Deep Think achieved the highest score in LiveCodeBench, one of the most challenging coding benchmarks. It also outperformed OpenAI’s o3 model in the MMMU test, which measures perception and reasoning abilities.

Currently, only trusted test users can access Deep Think through the Gemini API. Google hasn’t shared detailed information about how the feature works but noted that it is based on parallel thinking techniques. The company also stated that additional security testing will be conducted before the feature is made widely available.

In addition to Deep Think, Google announced an update for the Gemini 2.5 Flash model. This updated version brings improvements in coding, multimedia processing, long content comprehension, and logical reasoning. Google shared that the updated 2.5 Flash model is now available for trial use via AI Studio, Vertex AI, and Gemini apps, and it will be generally available to developers in June.

Google also introduced a new AI model called Gemini Diffusion, claiming it operates 4 to 5 times faster than similar systems while offering comparable performance to larger models. Gemini Diffusion is also currently available only to trusted testers.

You Might Also Like;

Copy URL
URL Copied