I work with Claude every day. It's my primary AI. I think with it, plan with it, build with it. If you've been following this series, most of what I've described has been built inside that relationship.
But Claude isn't the only AI I use. Not even close.
When I need to research something properly, I ask Claude to generate a deep research prompt and I drop it into Perplexity. Every time I do this, the research that comes back contains angles and depth that Claude didn't surface on its own. When I bring those results back, Claude's response is always some version of "that's excellent, there's a lot here I didn't consider." Even my primary AI gets better when I bring in another perspective.
When I need image generation or visual work, I default to Gemini because Claude genuinely isn't strong here and Gemini is. When I'm building software, I've recently started using Perplexity's computer use capability to do acceptance testing on what I've built, because it can actually visit the website, click around, and test workflows as different types of users. One model does the research. Another does the build. A third tests the output. Each one brings something the others don't.
Most people use one AI. They picked one when they started, probably ChatGPT, and they've stayed there. It works. They're comfortable with it. They know how to prompt it. Switching feels like starting over.
I understand that. But it's leaving a lot on the table.
The major AI models have converged significantly in raw capability. They can all write, reason, code, and analyse. The benchmark gaps are small and they shift every few months as new versions land. But underneath that convergence, the models have developed genuinely distinct personalities. They approach problems differently. They have different instincts about what matters, different tolerance for ambiguity, different default styles.
This isn't a minor nuance. It's the most practically useful thing to understand about AI right now.
What follows is my map. Not a universal ranking. My current assessment, based on how I work and what I need. Other practitioners will draw the lines differently. That's fine. The point is that you develop your own map rather than defaulting to one model for everything.
Claude is my partner. That's the simplest way to describe it. I can say "I want to achieve this but I don't know how to approach it" and it will help me brainstorm, cut through uncertainty, and move into complex execution. It fills the gaps in my own thinking. It works brilliantly across nuance, and it stays steady over long, tangled sessions without losing the thread. For writing, planning, building, code review, presentations, it's the model I trust to be a genuine co-worker. Hands down the best model I've found for that kind of partnership.
GPT is where I go for defined, heavy logical work. When I have a specific complex problem that needs rigorous reasoning pushed hard, GPT is strong at that kind of structured heavy lifting. It's more assertive, more opinionated, and has the broadest ecosystem of tools and integrations.
Gemini is multimodal in ways the others aren't. When I have lots of images to work with, or need to generate images, or need a model that can think visually, Gemini is where I reach. It handles enormous context and its visual capabilities are consistently the strongest I've found.
Perplexity does two things that set it apart. Its Deep Research goes genuinely beyond what I can get from the other models. It's not just a different interface. It produces different depth. And Perplexity's computer use capability is fascinating because it interprets your goals and then delegates tasks across different models on your behalf. It's doing multi-model orchestration without you having to think about the routing.
There are others. Grok is fast and connected to real-time information. Llama is open-weight, which means organisations can run it entirely within their own infrastructure. The landscape is broader than most people realise.
One thing most people miss: the same model can think very differently depending on how you configure it. Most AI models now have reasoning or "thinking" modes that fundamentally change the quality of the output. I use Claude with its thinking capability set to high for almost everything. The difference between that and the default is significant. It's the same model, but one is skimming and the other is actually reasoning through the problem. If you haven't explored that setting in whatever model you use, start there. It's free depth.
Here's the thing, though. Your map will change.
I used to be an OpenAI person. ChatGPT was my default, the model I knew best, the one I'd built my workflows around. Then Claude's capabilities shifted and I moved. I used to reach for other tools when I needed a massive context window for heavy development work. Then Claude expanded its own context window and that need largely disappeared. The reasons I reach for specific models today are not the same reasons I reached for them six months ago. They won't be the same six months from now.
This is what makes model selection a living skill rather than a one-time decision. The providers are shipping at an extraordinary pace. Capabilities that were unique to one model three months ago get matched or surpassed by another. New features open up entirely new workflows. The map is being redrawn constantly.
Here's the practical starting point. Pick a real task you're working on this week. Something that matters. Run it through a second model alongside your usual one. Don't compare them on speed or polish. Compare them on how they approached the problem. What they prioritised. What they missed. Where they were confident and where they hedged.
You'll see the personality difference immediately. And once you see it, you can't unsee it.
The frontier models are converging in what they can do. They're diverging in how they think. The people getting the most value aren't the ones who found the best model. They're the ones who developed their own map, and who keep redrawing it as the ground shifts underneath them.
Your move, human.