Wink Pings

Is Gemini 4 Neon Underrated? 5 Days of Testing, 10 Takeaways

Greg Isenberg got early access to Gemini 4 Neon and tested it for five straight days. Invoices, customer call recordings, local business websites—he threw everything at it. 10 takeaways, 1 conclusion: this model is underrated. Full video included.

Greg Isenberg got early access to Gemini 4 Neon and tested it for five straight days. Invoices, customer call recordings, local business websites—he threw everything at it. Conclusion: this is the most underrated model of the year.

Here are the 10 takeaways he summarized.

1. /swarm lets an AI team do the work itself
Type /swarm, staff it with a researcher, a designer, and a few engineers, point it at a directory or app in a niche, and let it run overnight.

2. The physics engine is strong enough for prototyping
The physics simulation is excellent. If you want to build a hardware product or a small game, have it build a prototype first, then decide whether to spend money.

3. Throw your company's messy invoices at it and get a CFO
Toss in chaotic invoices and have it generate a financial model. After that you can go back and forth with it like having an on-call CFO.

4. Local business websites hide cash-flow opportunities
Paste in a local business's website, and it will read the whole site and then build what they need: a calculator, a homepage. That's a business in itself.

5. Use games instead of PDFs
Build a mini-game or 3D configurator for a product and put an email field in front of it. It's way more fun than a PDF, and people will share it on their own.

6. A million-token context lets it dig up old promises
Put all of last quarter's customer call transcripts into Google AI Studio and ask it: what did we promise but not deliver? It can handle about a million tokens at once and stays sharp the whole way through.

7. Boring business chores are where it shines
Lead routing, CRM cleanup, invoice reconciliation. The work nobody wants to do, it does well.

8. Use Orchestrator mode day to day to save money
In Google's AI dev environment Antigravity, pick Orchestrator: Neon does the thinking and hands the typing work to the cheaper Hydrogen.

9. Turn thinking up, get stronger code
Set thinking to High and the same model scores three times higher on coding tests. Antigravity itself still needs work, but the results are already worth using.

10. It can write, but tell it not to use em dashes
Tell it to drop the em dashes and the writing quality is good. It defaults to short output—just say so if you want more.

Greg's takeaway: the model is genuinely cheap and works well. You can be bolder in trying out projects and automating tasks.

Full video on YouTube: http://youtube.com/watch?v=tCY7zMknCJQ

The video also includes community builds: Pelican dynamic product design, a robot dog moving across uneven limestone, and two pinball games with multi-ball physics. Plus 5 ways to make money and 4 things to watch out for.

发布时间: 2026-11-21 16:47