Google has presented Gemini 2.5 Its new artificial intelligence models (AI) designed to address “increasingly complex” problems, highlighting its first version Gemini 2.5 Pro Experimental which exceeds common reference points and demonstrates solid reasoning and programming capabilities.
After the launch of Gemini 2.0 in December of last year, the company has now released its new update of AI models Gemini 2.5to which it refers as its “smarter” model, With capabilities for think and reason your thoughts before answeringwhich translates into results with improved performance and greater precision.
Specifically, with Gemini 2.5 Google claims to have reached “a new level of performance” by combining a “significantly improved” base model with optimized possession. This makes models address more complex problems and supporting agents “even more capable and sensitive to context.”
Therefore, as the technology has shared in a statement in her blog, Google plans Integrate these thinking capabilities into all its models so that they can offer better results.
GEMINI 2.5 PRO EXPERIMENTAL
In this sense, Google has announced its First Gemini 2.5 Pro Experimental That, within the advances achieved with Gemini 2.5, it is its most advanced model for complex tasks, exceeding the main reference points for these AI models.
Thus, Gemini 2.5 Pro Head the classification of LMarena By a “significant” margin, which refers to a “highly capable and with a high quality style” model, as the company has stressed.
Following this line, this model also leads common programming, mathematics and science standards, thanks to its Reasoning capabilities. Specifically, according to the results shown, Gemini 2.5 Pro experimental It passes in Los Modelos Openi O3-Mini Y GPT-4.5 de Openai, Claude 3.7 Sonnet de Anthropic, Grok 3 Beta de Xai y r1 de deepsekat the main common reference points, which measure understanding, mathematics and programming, among other capacities.
Likewise, the technology has stressed that Gemini 2.5 Pro is based on the native multimodality and a wide context window. This version has a context window of 1 million tokensalthough soon it will be extended to 2 million, with a performance that improves that of previous generations.
In this way, you can understand great data sets and manage complex problems from different sources of information, regardless of whether data is used in the form of text, audio, images, video and even code repositories.
On the other hand, Google has also indicated that for the development of Gemini 2.5 they have focused on the Coding performancewith what they have achieved a “great jump” with respect to version 2.0. Thanks to this, Gemini 2.5 PRO stands out in the creation of Visually attractive web applications and agentic code applicationsin addition to the transformation and editing of code.
In fact, as Google has shown in an example video, Gemini 2.5 Pro can use its reasoning capabilities to Create a video gameproducing the executable code “from a only application line “.
As the executive director of Google Deepmind, Demis Hassabis, said in an X (former Twitter), Gemini 2.5 Pro is in this regard “An impressive avant -garde model,” Number 1 in LMARENA with a difference of 39 EL points and with significant improvements in multimodal reasoning, programming and stem.
Gemini 2.5 Pro is an awesome state-of-the-art model, no.1 on LMArena by a whopping +39 ELO points, with significant improvements across the board in multimodal reasoning, coding & STEM. You can try it out now in AI Studio https://t.co/lLpF8ToTVJ & @GeminiApp with Gemini Advanced https://t.co/bgjabz8O1u
– Demis Hassabus (@demishabisabis) March 25, 2025
With all this, the technology has indicated that Gemini 2.5 Pro is now available in Google to studyas well as in the Gemini application for users signed to Gemini Advanced. He has also advanced that he will soon come to Verex AI.