Hacker News Reader

Comment by rayiner | Hacker News Reader

I use both Grok 4.5 and Opus 5. They’re both very good and Grok is faster and cheaper.

Opus 5 is terrible. I'd even say it's a step backwards from 4.8. I'm getting high error rates from it, and then it catches the error, and then it sometimes errors the error fix (!).

Just today I had to switch another agent to Fable with the instruction, "Please clean up the mess that Opus 5 made, thanks"

The other day, Sol called Opus 5's handoff (a skill I have that is basically a compaction, but just written to a file not tied to one LLM) "incoherent", that was a new one.

Opus 4.8 or Fable (at great expense) are the only ones that aren't frustrating for me.

Every time when Opus 5 needs a design decision and presents me with suggestions/recommendations, I switch to Fable and ask it to think again, and it almost always replies something like "Actually my previous suggestions were wrong" and describes in detail a bunch of ways in which Opus 5's suggestions were indeed complete garbage.

Strange that i do not experience this. Its been great in my experience. But that may simple be because i switched from typing most of my prompts. To just dictating my prompts in a long and convoluted way and letting the LLM extra the information.

It allows for much more context that flow with your thoughts. Where as when you type, you tend to shorten you thinking process trying to get the bulleting points in, but that often ignores smaller things. And then you think "i can add this later", but that never happens because rabbit chasing the LLM.

So far all the suggestion that Opus 5.0 offered me, always aligned with what i wanted. Its not just Opus that i noticed this with.

The same happens if you ask Opus 5 to "think again"

Interesting. My experience has been similar. Opus 4.8 was awesome. Opus 5 feels a little off, although I can't put my finger on exactly what it is.

Well they say Opus was trained for the subordinate role, so it doesn't excel in global view of things.

It may be a good subagent but probably not a great decision maker.

Thanks, that’s interesting to know. I don’t know much about LLMs so I use 5 because it’s a bigger number than 4.8.

Same here. Regularly reverting back to Opus 4.8 after 5.0 being terrible.

Anthropic does this all the time (ruins their models for users) while they screw around with system prompts. Oh but it's for your own good of course! They know what's best for us all, if we would just give them a monopoly.

I can't wait until OpenAI/Grok/Chinese models surpass them enough that their main character syndrome and smug doomerism no longer draws much media attention.

Opus 4.5 gang here :)

I've reverted enough times I just pin this version.

Reply on news.ycombinator.com