In November 2022, we marveled at GPT-3.5 writing poems in the style of Victor Hugo. In 2025, Gemini 3 codes you a spaceship simulator in 90 seconds.
Only three years separate these two moments. And yet, the leap is staggering.
Models are no longer just “good at text”: they reason, plan and create, with a PhD level of intelligence, and a cost that is collapsing. Concretely, the cost of a million tokens has dropped a hundredfold in two years…
Gemini 3: the most advanced AI tool at launch, in November 2025
Available almost everywhere on Google AI Studio (the developer space):
On the Gemini app, or on Google's AI Mode (only launched in France in July 2026).
On performance, this is not just a new model. It is a show of force. At launch, it beat every benchmark:
That's great, but concretely what changes?
Vibe coding a 3D video game
You can take a look at Google AI Studio in the gallery section to see a range of creations by Gemini 3 that are all stunning.
Here for example, a game vibe coded with Gemini 3, inspired by Pac-Man but in 3D on a tiny planet.
I sincerely invite you to try it: you describe your game idea and the AI generates the 3D environment, the camera controls, the visual style and the assets… in a single flow.
Building an app in natural language
A simple request like “I want an app to manage my expenses” → the tool proposes a plan, builds the interface, handles the logic, asks me to validate, then compiles.
But what I find really useful is, for example, generating an interface to explain a concept. Here, for instance, understanding DDoS in cybersecurity, with a simple prompt “explain DDoS to me”:
Cloning an app from a simple screenshot
By uploading a screenshot, I can ask “Reproduce this interface with my colors and add a contact section”. Gemini 3 gets to work.
Here is an example with CapCut for video editing:
Coding without coding: why Antigravity deserves your attention
You don't code?
That's not the point. Because using a computer today is already coding without knowing it. And Antigravity starts from exactly this idea. The tool is comparable to Claude Code and Codex.
And even without being technical, it can be really worth digging into. Antigravity is an AI-driven work environment where every action can be delegated to agents, even complex ones, without ever opening a line of code.
What does it look like concretely?
- An agent inbox: you give them a mission (“sort my files”, “build a site”, “analyze this data”), and they run it in the background.
- A real control interface: you follow the progress, validate the choices, adjust as needed… just like with a colleague.
But what makes Antigravity different is its “agent-first” logic → it is not an improved dev tool, it is a new way of using your computer.
The day AI understands intent and drives the tools for us, the paradigm shifts. Antigravity is a bit like that moment.
Google keeps expanding its AI ecosystem
For several months now, Google has been blowing me away!
For me, they are catching up to ChatGPT… and nobody really talks about it. At the end of October 2025, the Gemini app had passed 650 million monthly users, against more than 800 million weekly users for ChatGPT.
It is not just a model, but a whole ecosystem, used at every level:
Search (AI Overviews, NotebookLM)
Workspace (Docs, Gmail, Meet, Gsheet…)
Creation (Veo, Nano Banana, Imagen)
Dev (AI Studio, Vertex AI, Antigravity,…)
The figures Google announced at the same time are pretty staggering:
- Veo 3 has already generated more than 230 million videos
- AI Mode = 75M daily active users
- Gemini = 3x more queries between Q2 and Q3
Not to mention Nano Banana Pro, released in November 2025, which I later tested on concrete image generation use cases. Its strengths:
- Visualization of complex information (educational infographics, interactive storyboards),
- Multilingual and typographic rendering that is precise,
- Advanced creative control (focus, lighting, text integrated into the image),
- Consistency maintained across characters or scenes (up to 14 combined elements!)
