Hacker News Reader

I’ll try 5.4 again but when I spent a couple of hours with it, I found it much weaker than Claude code + opus 4.6 for working on large projects.

Are you delegating substantial work like planning and executing refactors, or more at single-line and function-level work?

I think the gap is smaller than it has been in the past but I largely agree with you, generally larger work is done much better with Claude Code.

GPT 5.4 keeps erroring out when I’ve tried it on long complex tasks. Might be works harness I guess.

I'll do most of my work in opencode with GPT 5.3/5.4 and just spin up Claude Code for a review/cleanup pass.

Opencode just works faster and I prefer the UX.