Google showed a Gemini 4 Argon demonstration today in which the model generated 5,000 lines of math to create a sunlit 3D library, according to the company. Google says the work began with a brief prompt and used neither assets nor libraries, before the model produced a video explaining its process.
The company presents the exercise as an example of Argon handling extended assignments through reasoning and the ability to work across different types of input. Google says users can give the model research documents, unprocessed photos and projects requiring multiple steps, with Argon handling reasoning, programming and tool coordination.
Read more: Google Brings Gemini 4 Argon and Tasks to the Gemini App
Google also says Argon is taking on work internally that previously demanded months of manual effort. Its stated uses range from research to creating interactive designs in 3D, extending the pitch beyond the programming tasks highlighted in its earlier developer rollout.
Google began distributing Gemini 4 Argon through Android Studio, Google Antigravity and Google AI Studio on October 13, with access depending on a Google AI Ultra subscription or paid Gemini API use. The model also reached the Gemini app that day for Google AI Pro and Ultra subscribers, alongside task assignment capabilities that Google says can keep working while users are away.
