AI last week: Gemini 3, NVIDIA’s results and Nano Banana Pro
Google released Gemini 3 and an image model built on it, while NVIDIA reported record quarterly revenue.
This is an edited version of the weekly AI recap on the original blog, covering 17–23 November 2025.
Google began rolling out Gemini 3
On 18 November, Google introduced Gemini 3 Pro with multimodal reasoning. A useful test was to compare it on a real task with several inputs, such as reading a diagram and explaining how it relates to a written specification. Google's Gemini 3 announcement
NVIDIA reported record quarterly revenue
NVIDIA reported revenue of $57.0 billion for the quarter ended 26 October, up 62% year over year. Its Data Center business accounted for $51.2 billion. The figures show the scale of NVIDIA's sales at the time. To assess a specific buyer's return, we would still need to know how it used that infrastructure. NVIDIA results, 19 November
Google added an image model to the Gemini 3 family
On 20 November, Google introduced Nano Banana Pro, based on Gemini 3 Pro, for generating and editing images. The company highlighted text rendering and editing controls. For a team making diagrams or product assets, the useful check was whether labels and small details held up through revisions. Google's Nano Banana Pro announcement
The open and closed model trade-off
The original roundup asked how much control teams need over the models they use. The practical comparison is about where data can go, whether the model can be run or modified independently, and who maintains it. The answer can vary by workflow even inside the same company.