Why Google Claims to Have the 'Smartest AI Model'

Google is pushing the Gemini model to compete with rivals. Photo: Bloomberg .

After updating the models available to users, Google announced the Gemini 2.5 Pro beta for paid teams and Advanced developers on March 25. Like before, the search giant is doing a mid-year update with a significant model upgrade.

Notably, all the tools in the Gemini 2.5 family, including the future model, are products that are capable of thinking and reasoning through their own thoughts before responding to the user. This helps to improve efficiency and accuracy in generating information.

Additionally, Google is building models that can think directly and support other models to allow them to handle complex problems and increase their contextual awareness.

Compared to Flash Thinking 2.0, which was first revealed in December 2024, Google no longer labels it as “Inference” so casually. Users can now use the new Show Thinking feature in the Gemini app to see the AI’s train of thought.

"In the field of AI, a system's ability to reason is not just about classification and prediction. It also requires the ability to analyze information, draw reasonable conclusions, incorporate context and nuance, and ultimately deliver useful information to users," Google wrote on the blog.

According to 9to5Google , Gemini 2.5 achieves good performance by combining a significantly improved base model with deep training. Google says it topped the LMArena leaderboard by “a significant margin over many competitors.” It also outperformed many other models in math (AIME 2025) and science (GPQA diamond) benchmarks.

"Gemini 2.5 also scored 18.8% on unengineered models on Humanity's Last Exam, a test that uses data designed by hundreds of experts in the field of artificial intelligence. It is also used to capture the boundary between human knowledge and human reasoning," Google added.

Source: https://znews.vn/google-ra-mat-mo-hinh-ai-thong-minh-nhat-post1540993.html