Google DeepMind has released Gemini Ultra 2.0, and benchmark results show it surpassing GPT-4o on 8 of the 10 most widely used AI evaluation tests. The results end a period of relative AI stagnation at Google and signal that the company's massive investment in AI infrastructure β over $75 billion in 2024 alone β is paying off.
Gemini Ultra 2.0's most notable improvements are in multimodal reasoning: the model can analyze videos, images, audio, and text simultaneously, making connections across formats that previous models could not handle. In a live demonstration, it watched a 20-minute lecture video and produced a detailed study guide with chapter breakdowns, key concepts, and practice questions in under 30 seconds.
The model is being integrated into Google Search, Gmail, Google Docs, and Google Meet starting this quarter. For the 3 billion people who use Google products daily β including over 250 million American users β Gemini Ultra will become an ambient AI layer woven into everyday digital life.
OpenAI CEO Sam Altman acknowledged the results on X, writing: "Congrats to the Gemini team. This is what healthy competition looks like."
