James Dooley: ChatGPT versus Claude versus Perplexity AI. Today I am joined with Dennis Yu, who has done a lot of split testing between different LLMs. He works with contractors on knowledge panels, local SEO, maps and rankings. So today, based on your testing around pricing, context windows and performance, which one comes out on top and why?

Dennis Yu: Let me show you instead of just giving an opinion. Most people do not show their actual workflows. So with ChatGPT, you have things like Atlas, which is a Chromium-based browser. I can run tasks while browsing, open side panels and tell it to execute actions across tabs, emails and tools.

I have spent around $20,000 over six months using ChatGPT agent mode. It has been strong for launching projects, opening tabs and doing multi-step actions. But it still has limits. It times out and struggles with longer execution.

Now with Claude, the difference is persistence. Claude can run for 40 minutes, 1 hour or longer. I have seen it take over 100 steps in a single workflow. I have pushed it to 700 steps running for 4 to 5 hours.

For example, I ask it to analyse everything we have done, create articles, document processes, take screenshots and improve SOPs. It builds recursive systems where agents learn from previous outputs. That is where Claude stands out.

James Dooley: So are you doing everything in Claude now? No ChatGPT or Perplexity?

Dennis Yu: Not fully. Perplexity AI still plays a role. It has its own browser and allows model switching. That means I can still use Claude models through Perplexity when I hit limits.

Perplexity is useful for handling uploads, downloads and switching models quickly. If I hit usage caps in Claude, I move to Perplexity and continue using the same model. That is how I keep costs low while still consuming huge compute.

Right now, I am probably using $50,000 worth of tokens while paying only a few hundred dollars monthly. That only works because of how these platforms price access.

James Dooley: So if someone is just starting, what should they do?

Dennis Yu: Start simple. Use Claude Pro at $20 per month. If you hit limits, move to $100, then $200. Only scale when you actually use it.

Focus on repetitive tasks first. Reporting, audits, content repurposing. Upload past examples so the system learns your workflows. Once that is in place, you can scale output massively.

My workflow has changed completely. If I think of something, I just tell the system and it gets done. There is no backlog anymore.

James Dooley: And for something simple like content repurposing or social media posting, how easy is it?

Dennis Yu: Very easy. Let me show you. Inside Chrome with Claude, I can open a page and dictate instructions. For example, I can tell it to:

Take existing YouTube content. Turn it into blog posts. Follow SEO guidelines. Add tags, categories and embeds. Create a meta article explaining the process. Estimate token cost and human time. Then repurpose snippets into social posts. Publish them. Message team members with results.

Once I give that instruction, it runs for hours without intervention.

James Dooley: That is insane. Anyone watching this can clearly see the difference. Most people talk theory. You have shown actual execution.

Dennis Yu: Exactly. Things change fast, but right now Claude for Chrome is the most powerful for doing real work.

James Dooley: Appreciate it. Anyone watching, check the links for more breakdowns. Dennis goes deep on all LLMs and how to use them properly.