Google 发布 Gemini 4，主打编码、金融与 3D 推理

以下是视频口播的自动转写，可能有错字。

Google's gemini 4 is finally live a few hours back. Google launched gemini 4 argon their most advanced model yet. It's built for deep complex work across coding finance, legal cybersecurity and science.

It's ranked number one on the valves index, a benchmark that measures how well ai performs real economic work like finance coding legal and tax.

It also scored 87% on the embodied reasoning benchmark, which tests how well a model understands physical space and 3d environments.

Google says that beats every other frontier model out there. Here's what they've actually done with it. Inside google, it helped the chrome team untangle a 6 million line chunk of legacy code.

And finish months of work in just weeks. It made google's own gemma models almost 3 times faster in days. It took a single image and drove blender on its own to sculpt a full gothic 3d scene, built a detailed pagoda temple complex with lighting and foliage, and even turned photos of a real desk into a playable racing game.

It watched a raw video of a human playing cornhole, figured out the movements, wrote the simulation code, and a real humanoid robot then sank 7 shots in a row.

You can use it right now through the gemini api in google ai studio. Inside google anti gravity, and in the gemini app and ai mode in search, if you're on google ai pro or ultra.

And that's it.