Development with GPT 5.4

I just finished rewriting my blog site with openai's new model GPT 5.4 through opencode with github copilot. The results are...wow. To have this level of code delivered for a very low usage of copilot's premium request for the month is crazy to see. Originally this website, the react tic tac toe project and the Add-Non-Steam-Games projects were all meant as projects to test how good the model was. I've used 5.2 codex to help in building a blog before and I never got this level of quality. I thought it was possible that needing to rewrite a project I had originally made with another ai model slowed it down but I gave openai's 5.4 the old version of this site and it was able to rewrite it all in one go, only needing to ask one question to build the entire base of the project. I'm aware this project is smaller than the other blog I am working on but it is still really impressive. These 3 separate projects from top to bottom took only 17% of my github copilot premium usage. Another really impressive part of 5.4 seems to be it's tool usage. It does a very good job of fetching the information it needs to know how to use different parts of a stack

My process

Take a master prompt I use for projects (I have separate powershell, webdev and rmm prompts), have opus 4.6 rewrite that specifically for each project then give that master prompt to gpt 5.4

Lasting Impressions

I'm left in awe of this new model and I'm curious how well it compares to sonnet 4.6 if given the same tasks. I've mostly used opus for building websites or powershell projects like this because I wanted the model to be smart enough to think through what I threw at it, troubleshoot and test as needed without interruption or while asking clarifying questions but gpt 5.4 has changed my whole view of this. I've either underestimated models that aren't heavy weights like opus or gpt 5.4 is just that good in terms of code quality and efficiency.