Article ·
My Chat GPT-5 Breakdown (and which LLM I'm excited to start using)
I’ve logged a LOT of time with Chat GPT-5 (pro, thinking, instant, and auto) over the past few weeks.
The most important thing to remember with GPT-5 is that it isn’t ONE model – it is a router for SEVERAL models, and uses contextual clues to identify which model is should engage at any given time.
Unfortunately, this:
- Still hasn’t been explained well and
- Still feels like 3 cost optimizations in a trenchcoat as opposed to several models trying to give the best output to the user
With that said, GPT-5 Thinking and GPT-5 Pro (especially Pro once you figure out how to elicit the type of response you want) generate solid outputs–including code. I can tell a MASSIVE difference between GPT-o3/4o code projects and GPT-5 Pro code projects.
But I still am not having the types of conversations with GPT-5 that I had with previous models.
It is eager to ship an answer in a single turn. In the before times (you know, way back in July of 2025), I could flesh out ideas and build on them incrementally during conversations that felt collaborative. Now, it feels much more like a Q&A bot as opposed to a thought partner I’m engaging with to ideate and build.
Today, I’m working on a REALLY fun and hopefully helpful project (more to come on this soon). I’ve been using Claude Code daily for about a week, which made me realize I should check out regular ol’ Claude again.
WOW. I’m glad I did.
I’ve had two conversations with Sonnet 4.0 today. One about the project, and one about developing some ideas and materials around agents. It felt like I had my thought partner back. It was fantastic.
In particular, I’m a big fan of Claude’s styles (learning, explanatory, concise, etc.). They really help tailor the outputs to what I need in the moment.
If you’ve slept on Claude, now’s the time to check it out again.
PS – For good measure, I did some image generation and editing with Gemini today. That earned another “WOW” from me. Be sure to check it out.