Original video in Chinese.
Key Takeaway
- GitHub Copilot Pro+ is my main setup: because Claude Code keeps getting banned due to regional restrictions and has become unstable, I switched to OpenCode + the oh-my-openagent plugin. Subscribing to Copilot Pro+ lets me call multiple models, which is both flexible and hassle-free, and it’s currently the most reliable base for development.
- Perplexity Max is for serious work: in scenarios that demand high accuracy and rigor, such as investment decisions or work reports, Perplexity is far better than general-purpose AI. It pulls authoritative financial APIs in real time, forces source citations, and uses sub-question decomposition plus cross-validation. Perplexity Computer (Agentic Workflow) and the newly launched Personal Computer (cloud lobster) make it a heavy-duty research tool.
- Grok + Gemini are ecosystem supplements: Grok costs $20 a month, mainly for its 4-Agent orchestration, real-time X platform data, and web-reading capabilities; Gemini also costs $20 a month, relying on deep integration with Google’s entire ecosystem, especially the two-way connection with NotebookLM. I switch between them as needed and use whichever is stronger. The total cost is about $300 per month, which is completely acceptable.
I currently subscribe to four AIs. I’ve posted about them in Zhishi Xingqiu before. In this video, I’ll explain why I subscribe to them and give everyone a reference.
First is GitHub Copilot Pro+.
Why subscribe to Copilot?
Because I can use Copilot in OpenCode to call models from different providers.
Why not use Claude Code?
I’d like to, but they won’t let me.
Anthropic’s countermeasures are just too many, and it’s not only IP-based. Look, now they’ve added identity verification too.
I also said in Zhishi Xingqiu that the signal is already very clear: Anthropic will use every possible means to block users from unsupported regions.
So everyone should stop messing with it. There’s no way it can remain stable long term. If you suddenly get banned one day, it affects both your work and your mood. Why bother?
I switched to OpenCode several months ago. With the oh-my-openagent plugin, development has been completely fine, and it is absolutely no worse than Claude Code.
My newtype OS is also based on OpenCode. Its core is actually just a plugin. For convenience and easier configuration, I deliberately integrated OpenCode into it.
So which model do I use in OpenCode?
Anthropic and Google are clearly not supported. OpenAI and GitHub have officially announced their partnership. So if I want more options, subscribing to GitHub Copilot is the only choice.
I’m currently subscribed to the Pro+ tier. I used the Pro tier before, and the quota was nowhere near enough. Honestly, Pro+ still isn’t quite enough either. But fortunately, once you run out, you can add your own budget and keep using it.
Although the models provided by Copilot are somewhat limited in context window, overall, you don’t really have a better choice, do you?
Next is Perplexity Max. I’ve recommended this tool more than once in Zhishi Xingqiu.
If you’re doing serious work, like submitting a work report or making an investment decision, then you definitely need Perplexity. Its accuracy and rigor are absolutely stronger than products from peers like OpenAI and Google. Some friends in Zhishi Xingqiu who do quant work started using it after my recommendation, and all said it’s very good.
Besides search, I also use Perplexity Computer a lot in my daily work.
For example, when the U.S. military action against Iran happened, I had it do a comprehensive information gathering job and continuously monitor the latest developments, because this affected my investments.
On the development side, I’m currently building an automated trading bot for prediction markets. I’ll have it gather information and then provide development suggestions. After that, I’ll take the Markdown document into OpenCode to make decisions and execute.
Just yesterday, the Personal Computer feature was finally launched. It’s Perplexity’s version of the lobster. Local operations and cloud model invocation are now connected. I’ll make a video once I’ve tried it.
So you can see that one Max subscription lets me do so many things. That $200 is truly worth every penny!
Finally, let’s talk about my Grok and Gemini subscriptions together.
There aren’t many people in China using Grok. But this Musk AI is actually very well made. Especially this 4.2 update, which added multi-agent orchestration, with four specialized agents working in parallel. They handle:
Task decomposition and result synthesis, real-time search and fact verification, math/code/logic reasoning, and creativity and writing.
I’ve been using it for a while now, and I’m very satisfied with this multi-agent orchestration. In addition, Grok can access unique X platform data, and it can also read web pages like GitHub, which is very convenient.
As for Gemini, to be honest, I’ve used it less recently—because it’s gotten dumber. But I’ll still keep subscribing, because Google’s ecosystem integration is just too good.
For example, the two-way integration with NotebookLM has gotten even better recently. In particular, all Gemini conversations now include a new option to add them to a notebook, so they can continue as new material for NotebookLM research. That is my absolute favorite change; it’s really super practical!
Also, the newly released Mac version of Gemini is pretty good. Although it doesn’t have many standout features, it at least fills in a weakness. These are all good things for longtime users.
I’m subscribed to the $20 tier for both Grok and Gemini, and that’s enough. Over a period of time, I just use whichever one is stronger more often.
These four subscriptions together cost me a little over $300 a month. For me, the cost is completely acceptable. If anything changes later, such as a new AI tool subscription being added, I’ll mention it in Zhishi Xingqiu anytime.
OK, that’s it for this episode. If you want to learn AI, want to become a super individual, and want to find like-minded people, come join our newtype community. See you next time!