The new Gemini 3.1 Pro is out. I asked it to write a post about itself and create an image for you. I asked it to impress and joke (well...). For a Friday evening, it's fine:
"So, bros and researchers, they just dropped the new Gemini 3.1 Pro, and this thing is legit.
Straight to benchmarks: it holds its ground on evals. Drops scores that clearly show the reasoning works. The model genuinely gets the gist and runs hypotheses, not just outputs memorized stuff. Crunches millions of tokens of context without choking.
If you're on the paid tier on mobile, they shipped Gemini Live mode. Chats in real-time, you can interrupt freely. Share your camera or screen — it'll calmly break down what's going on in your app.
On the generation side:
- Images: Nano Banana under the hood. Renders text without errors, you can iteratively improve the image in dialogue. Generous limit — a thousand generations per day. But no photoshopping politicians, there are filters.
- Videos: Runs through Veo. Outputs high-res with native sound, you can even set the first and last frame. But quota is tight — three videos per day.
- Music: Lyria 3 makes 30-second tracks beautifully, but Google slaps SynthID watermarks, so you can't pass it off as your own beat."
And the main feature: if you don't have subscription money, head to Google AI Studio. You can try and test this beauty for free, strictly for developers.