My experience with the new GPT-5.5 model
Yesterday, OpenAI launched the newest model GPT-5.5 which I was too excited about for the last few days. And here are my experiences so far with using the new model inside the Codex app.
- GPT-5.5 is far better in frontend design than other OpenAI models like GPT-5.4 or 5.3-Codex. But it's still not at the level of Opus 4.7 of 4.6 when it comes to UI design.
- Even though it was claimed by Sam Altman that the new model uses fewer tokens per task than previous models, it's not the case when I use it. In my case, it used way more tokens than GPT-5.4 because earlier I could never finish the 5-hour limit on the $100 plan, but today I did, as you see below.
- The new model, GPT-5.5, definitely feels slightly faster than the previous, GPT-5.4, model in the standard mode. I think, both are equally faster in the speed/fast mode.
- I didn't notice any significant improvement in the programming though with the GPT-5.5 model, it feels similarly capable as GPT-5.4 model. Or, maybe I haven't tested this enough yet.
Also, the new Codex app feels snappier and the UI and UX is slightly improved now.
Apart from this, I can still not rely completely on the Codex subscription itself. I can use OpenCode with models like GLM-5.1 and Kimi-K2.6 for frontend design and then GPT-5.5 for everything else. The OpenCode Go plan is very generous at $10/mo and has crazy limits as well. So if I had to reduce my monthly spending, I would go with OpenCode Go ($10/mo) + ChatGPT Plus ($20/mo) and maybe $20-30 extra on OpenCode, and that should cover all my requirements.
I'm still playing with the GPT-5.5 model so if I find something new worth sharing, I will keep this post updated.
Webmentions