After providing a fresh “open” AI model with better performance on one GPU, Google has now introduced an update of AI models for its products from Gemini 2.5which combines a “significantly improved basic model with improved after training” for better overall performance. He claims that the first edition, Gemini 2.5 Pro Experimental, conducts competition on the part of OPENAI, Anthropic, Xai and Deepek with joint AI reference points that measure understanding, mathematics, coding and other possibilities. The fresh model is available for access Google to learn Or for advanced Gemini subscribers in the developed application model.
The company also advertises the native multimodality of Gemini as an advantage, because it is able to interpret not only the text, but also sound, images, video and code, and says that 2 million token contextual window is “soon” to aid him process more data. Demis Hassabis Google Deepmind called Gemini 2.5 PRO “Amazing Most current model, no In the post on x.
Google claims that it has jumped into quality, because Gemini models are now “reasoning” models that process the task step by step and make more conscious decisions, which, they say, results in better answers and answers to complex hints. Now the blog post is: “… We are building these thinking opportunities directly in all our models so that they can deal with more intricate problems and support even more talented, conscious contextual agents.”
One Video demo It shows 2.5 Pro using these reasoning to program a video game based on one poem: