Google rolls out Gemini 4 Argon, its most advanced AI model

Google just released Gemini for Argon, what it's calling its most advanced AI model with major improvements in coding, cybersecurity, and complex professional work. Now, Gemini 4 was expected back in May as part of their IO developer conference. It was delayed as Google chose to focus on faster, cheaper flash models before making a major push at the frontier. Now, Google tells me the model set a new record in real-world software engineering, ties with rival models from OpenAI and xAI for first place in cybersecurity. Then leads another benchmark measuring performance across finance, law, and other professional work. Now, Google telling me a short time ago that they're already using Argon internally for the past few weeks, including to optimize its data centers, potentially freeing up hundreds of terabytes of memory without having to spend more capex dollars on additional hardware. But, as with many other cutting-edge models, Google is initially limiting access to select cybersecurity defenders and some enterprise cloud customers while working with the US government on pre-release safety evaluations. This, of course, coming a day after CEO Sundar Pichai signed President Trump's voluntary AI safety accord at the White House. And as Sam Altman tells CNBC, OpenAI is holding back its latest Astro model over safety concerns.