I didn't even know GTP-5.3 Codex Spark existed, and I happened to try it out for the first time because I ran out of tokens.
It's incredibly fast; when you give it a command, it almost instantly changes all the code. There are two problems, though. Its intelligence is a bit lacking, so it starts working without thinking. So even if you talk to it like you would with 5.5, it doesn't understand. It interprets everything literally and implements it, so there was even a time when it deleted my database once (Ugh... I still get angry just thinking about it). It's so fast that there's nothing you can do about it. (You're already gone...)
The other problem is that the context window is incredibly small. If the conversation gets a little long, it keeps compressing, and when it compresses, it forgets recent things, so I've even gotten stuck in an infinite loop. "What was I trying to do? Ah, yes, I was going to do that. I will... (Context compression) What was I trying to do? Ah, yes, I was going to do that. I will... (Context compressing...)"
So I thought it wouldn't be usable for general purposes and put it aside, but then I found a use for it: document summarization. In terms of summarizing documents, the quality is better than deepseekv4 flash and it's faster. Deepseek was also very cheap, but Codex Spark was just lying around, so I can use it without any burden. I've used about 12% in the past week, and I feel great.ㅎㅎ
I'm curious how other members are using it.
