Google's next flagship AI model is currently being tested internally using a system called Antigravity while engineers finalize safety guardrails. Google DeepMind chief Koray Kavukcuoglu told The Information that Gemini 4 is in early post-training and will be released as soon as possible, with the full model arriving well before the end of the year. The company intends to iterate rapidly on an early version rather than waiting for a fully matured model checkpoint.

The accelerated timeline aims to close the performance gap with competitors Anthropic and OpenAI. Google has effectively moved past the development of Gemini 3.5 Pro to focus on smaller Flash models to speed up its learning cycles. Furthermore, Gemini 4 development is informing the design of the next 2 or 3 generations of Tensor Processing Units as Google sells its chips directly to compete with Nvidia.