Original video in Chinese.
Key Takeaway
- Gemini 3 is the strongest overall: the best model on the planet right now, with a historically huge leap in frontend capability (it can recreate UI/interaction one-to-one from a single sentence), completely crushing 2.5 in logic, content, and visualization, with extreme value for money at the $20 membership tier.
- The best programming combo: hand frontend work entirely to Gemini 3, and use Codex for backend, bug fixing, and architecture adjustments (the GPT-5-High $200 tier is the most reliable), completely saying goodbye to frontend suffering.
- The Software 3.0 era: Gemini 3 + Generative UI realizes Code is Intent. Intent is code, instantly generating fluid interactive interfaces, used once and thrown away; from 1.0 (code is logic) → 2.0 (code is data) → 3.0 (code is intent), Google’s full stack is the best, and it’s worth going heavily long.
Let me share my view of Gemini 3. I think that, overall, it is currently the best model on Earth.
If your need is AI coding, then I strongly recommend the combination of Gemini 3 and Codex.
You can feel completely comfortable handing frontend work over to Gemini 3. For the first time in history, AI can reach the level of human programmers on the frontend.
In the past, no matter how powerful the model was, it was always terrible at frontend work. I’ve used everything from Claude 3.5 to 4.5, and then GPT-5, and even something as simple as changing padding or adding auto-load when scrolling to the bottom was honestly a nightmare. It was extremely painful.
No matter how precisely you described it, those models still couldn’t understand your real needs. Even if you were very specific, they still couldn’t change it properly. You could only keep rerolling.
But now, with Gemini 3, you can give it one sentence and it can produce something that’s nearly one-to-one accurate. If you don’t believe me, you can try it in Studio, or look on X, where there are tons of examples.
As for the backend, after actually testing it, I found that Codex is still the most reliable. Really, that’s just the power of the GPT models. If you want to modify bugs or adjust the architecture, choosing Codex is absolutely the right call. I’m on the $200 tier now, and it feels amazing.
If you don’t have coding needs, then Gemini 3 may be the best choice for you right now. As I said in Knowledge Planet:
Even if you don’t use it for development, you can still use it to generate visual, highly interactive pages, or long images and posters. These can all be very helpful for improving your efficiency in processing information. I even saw someone use Gemini 3 to make a game, so kids could learn while playing.
In addition, Gemini 3’s logical ability and content ability are stronger than 2.5’s. Just discussing problems with it, asking it to help me organize logic and add missing information, is already worth far more than the $20 monthly fee.
So Gemini 3, along with the Gemini membership subscription, is definitely the best choice for most people.
If you are an investor, then after Gemini 3’s release, you really should pay close attention to Google stock. In fact, the market has already started shifting its attention. People have suddenly realized that Google is the strongest company in AI.
Think about it: they have TPUs, so they don’t need to rely on Nvidia; they have Gemini models, so they don’t need to rely on OpenAI or Anthropic; they have their own data centers and cloud services, which they can use themselves or rent out as compute; and they have the most comprehensive, complete application ecosystem, which lets AI actually land in the real world.
That’s what a “full-stack AI” really is!
So while U.S. stocks have pulled back over the past two months, I’ll probably buy some Google.
Finally, besides launching Gemini 3, Google actually launched one more thing that hasn’t gotten much attention: Generative UI.
Simply put, in the future, when you ask a question, AI won’t just give you a text answer. It will also generate an interactive interface to show dynamic effects.
Most importantly, these interfaces are not predesigned templates, but something AI generates on the spot.
If you combine Gemini 3 and Generative UI, you’ll see that we are getting closer and closer to Software 3.0.
In the Software 1.0 stage, Code is Logic.
This is the traditional software development model we are most familiar with. Programmers write the steps for solving a problem, step by step, as instructions that computers can understand.
The limitation of Software 1.0 is that every line of code and every piece of logic must be written in advance by humans. If it isn’t written in, or if it can’t be described by rules, then the program can’t run.
In the Software 2.0 stage, Code is Data.
Programmers no longer write code logic. Instead, they write the “objective function” and provide massive amounts of data. Neural networks automatically find solutions through optimization algorithms. The final “software” is not source code, but weight parameters.
The limitation of Software 2.0 is that although it solves perception problems, it is a closed black box and makes it very hard to carry out complex interactions with the external world.
The launch of Gemini 3 and Generative UI this time gave me a huge insight: I feel like we are already standing at the door of Software 3.0. In this stage, Code is Intent.
We only need to describe our intent and express our ideas. After thinking, the Gemini 3.0 model writes a bunch of code, runs it in a temporary environment, and ultimately presents it through a streamed, generated interactive interface.
In other words, by the 3.0 stage, the entire software stack, including the UI, becomes a kind of “fluid.” Software is the real-time fulfillment of intent, instantly generated by AI and used once before being thrown away.
In the Software 3.0 architecture, we no longer distinguish between frontend and backend. Instead, it becomes a relationship between “server-side decision-making” and “client-side rendering.”
Code is Cheap, Intent is King.
Welcome to the new era.
OK, that’s it for this episode. If you want to understand AI, want to become a super individual, and want to find like-minded people, come join our newtype community. See you next time!