Google Unveils Gemini 2.5 Pro, Its Smartest AI Model Yet, Beating OpenAI o3

Google has announced the Gemini 2.5 Pro AI model, calling it “its smartest model yet.” The neural network is part of the Gemini 2.5 family and outperforms previous versions in data analysis, programming, and solving complex problems, supporting context for up to 1 million tokens.

Image source: Google

The key feature of the Gemini 2.5 Pro, like all models in the Gemini 2.5 family, is the ability to reason, visualizing its thought process before giving the user a more precise and final answer. Unlike the previous generation of models (Gemini 2.0 Flash Thinking), Google no longer uses the Thinking label or displays the reasoning process. However, as 9to5Google points out, users can manually activate the “think out loud” feature to see the bot’s thought process.

Overall, Gemini 2.5 Pro showed a significant jump in performance thanks to an improved base model and post-training tweaks. Google notes that this version topped the LMArena rankings, which evaluates models based on user preferences, and also showed better results in math (AIME 2025) and science (GPQA diamond).

At the same time, in the Humanity’s Last Exam test, which is created by experts to test the limits of artificial intelligence in the field of knowledge and logic, Gemini 2.5 Pro achieved a record 18.8% without using additional tools. The model also received significant improvements in programming, especially in creating web applications and editing code.

In the software development space, Gemini 2.5 Pro scored highly on the SWE-Bench Verified benchmark, scoring 63.8% using a dedicated agent approach. It also has built-in multimodality, handling text, audio, images, video, large data sets, and even full code repositories.

The model’s context window offers a size of 1 million tokens, and in the near future it will increase to 2 million. In the next few weeks, Gemini 2.5 Pro will appear in Vertex AI, and later Google will introduce a pricing policy that allows using the AI ​​model in large-scale projects. For now, the model is available to paid subscribers and developers in test mode.

admin

Share
Published by
admin

Recent Posts

SnowRunner creators’ ‘revolutionary’ RoadCraft simulator earns ‘mixed’ reviews on Steam release

As promised, the “revolutionary” construction simulator RoadCraft from Saber Interactive (SnowRunner, Expeditions: A MudRunner Game)…

17 hours ago

Google has taught Meet to translate speech on the fly while preserving intonation and tone of voice

Google unveiled a new live translation feature for its Google Meet video conferencing service at…

17 hours ago

CMF Phone 2 Pro Review: Still Surprising

Last year, Nothing introduced the first smartphone under its budget sub-brand CMF by Nothing. The…

17 hours ago

Google Chrome Will Start Automatically Changing Weak or Hacked Passwords, But Will Ask for Permission First

At Google I/O, the company announced a new feature in Chrome that will automatically update…

17 hours ago

The End of Silent AI Video: Google Unveils Veo 3, the First Video Generator with Sound

Google presented the latest AI model for generating videos from text descriptions, Veo 3, at…

17 hours ago

GTX 750 Ti is no longer enough for the game: Ubisoft announced the system requirements of Rainbow Six Siege X

Publisher and developer Ubisoft has revealed the system requirements for Tom Clancy's Rainbow Six Siege…

2 days ago