ChatGPT 5 - A very good model, but not as hyped.

A couple of days ago, I posted about the Claude Code trial and not 12 hours after publishing, OpenAI dropped ChatGPT 5 that had at least one of my regular go-to YouTube channels make a statement about how amazing this new release is. I've also been meaning to try out Kilo Code, so I installed the extension, claimed my $20 free credit and put ChatGPT 5 to work.

The results... were not as surprising as comparing AI models from 2 years ago to Claude Code and the Sonnet model, but still a very good model from what I can tell. I just feels in some ways a minor increment at best, but more just "different". To test it (and Kilo Code) out, I took the same REQUIREMENTS.md for my previous projects and let Kilo at it, by using some simple prompts (e.g. Read, create a Plan and implement a specific directory).

Observations

• The combination of Kilo Code and the ChatGPT 5 model feels glacially slow compared to Claude Code and Sonnet.
  
  
• Being able to switch profiles out of box for Architect / Code / Debug was a nice touch, but in these smaller projects would understandably have limited value in having those pre-canned roles do anything specific. Architect does create some nice mermaid diagrams though.
  
  
• From a Kilo Code perspective, the UI showing how many tokens are in use was a nice touch, with an ETA on cost.
  
  

  

• Despite the glacial pacing, Kilo Code was taking screenshots every step of the way.
  
  
• But... despite the screenshots taking place, it appears that Kilo Code or the ChatGPT model completely ignored any images held within the markdown files. 👎👎👎
  
  
• When I did feed it a Layout directly as an attachment, ChatGPT did a pretty good of implementing the wireframe but I feel the Claude Code example wins out ... slightly.

Conclusion

There's a lot to like about ChatGPT 5 and it's pricing, but I can only describe my first impressions as follows:

Had I used ChatGPT 5 + Kilo Code first, and then discovered Claude Code - my opinion might be that Claude Code is a slight improvement on the Kilo Code model - especially around the ease of planning and significant boost in speed, but I wouldn't have said that it was a ground-breaking upgrade - and possibly inferior in some parts. Having used Claude Code first though, ChatGPT 5 does not - in my early usage - feel like it's living up to the hype that even Theo speaks about in his video, and that other designers are showing with their freeform prompts.

It's a very good model, but happy to be proven wrong as I continue to play.