On July 21, Google launched Gemini 3.6 Flash, highlighting speed and cost efficiency rather than raw AI power. Despite this, the model ranked only 10th on an independent AI benchmark, trailing behind rivals.
Google's approach is diverging from competitors like OpenAI and Anthropic, who chase recursive self-improvement AI systems that can build increasingly better versions of themselves. Instead, Google is investing in AI that grasps real-world physics and interactions, a concept known as world models.
DeepMind’s initiatives include Gemini Robotics and Genie 3, designed to understand physical environments and control machines. Recent expansions like Project Genie’s extension into Street View and SIMA 2, an AI agent learning through virtual 3D play, show this direction clearly.
Google’s CEO Sundar Pichai emphasizes personalized AI agents over leaderboard rankings. This shift comes amid some investor unease; Alphabet’s stock dropped 6% in June after top researchers left for rival firms.
While bigger models like Gemini 4 and Gemini 3.5 Pro are still in development, Google's clear bet on embodied AI marks a strategic break from the race to build the most powerful language models.
This material is for informational purposes and not financial advice.
