← Back to context

Comment by Tiberium

8 hours ago

It is absolutely a product of the model, GPT-5 behaves like this over API even without any extra prompts.

I prefer its personality (or lack of it) over Sonnet. And tends to produce less... sloppy code. But it's far slower, and Codex + it suffers from context degradation very badly. If you run a session too long, even with compaction, it starts to really lose the plot.