Claude Opus 5.5

I find Opus 5.5 is better at giving a high level adversarial "should you do this to begin with" where GPT-6.1 is happy to go down any wrong path.

Opus 5 was the opposite where you give it a big list of tasks, does about 10-25% of it then drops out early explaining why it stopped instead of just investigating the complications it found on the way.

Opus 5.5 (O5.5) has been exceptional for me in Claude Code over the last six days. Architecture-first, DRY/SOLID coding out of the box, exceptional communication style, phenomenal token efficiency. I’ve been working with it basically nonstop, ~12 hours a day, since last Wednesday. But right around the time my monthly limit reset this evening, I noticed an extreme shift in communication style and coding behavior that feels suspiciously like Opus 5 (O5). For the last week, O5.5 would take my requirements and immediately get to work, usually knocking out a feature in minutes and doing an incredible job across implementation, UX, architecture, and token efficiency. Tonight, I’m seeing something very different.

Where O5.5 consistently respected DRY principles and the existing architecture, the modules created tonight are suddenly happy to reinvent every wheel that came before them. The difference in ability is not subtle. The other tell is token usage. O5.5 has been startlingly efficient compared with O5, which was an absolute token-eating monster. I’ve been getting substantially more feature work done with O5.5 at a fraction of the token usage. Tonight, simple tasks and relatively small features are suddenly devouring tokens at a pace that feels much more like O5. I’m especially sensitive to this because I have to request an increase in my spend limit every time I hit it. On O5, I was sending a request almost every day. On O5.5, I hadn’t needed to send a single one all week. Then tonight I jumped from roughly 70% to 90% usage in about an hour.

The broader complaints line up: Sol 6.1 is slow and quota-hungry, while Opus feels faster and lets people work longer before the usage meter starts screaming.

SOL 5.6 and 6.1 both are monsters for high level backend coding.

I just tried switching to the "accept edits on" mode, i am still getting the same push back.

I would manually log into my online banking, I would let Claude select the payees and then insert the amount to pay via playwright or chrome connect. I would be the one authorising the payment, but with the Opus 5.5 And a few other versions that I've tried, including Sonnet it seems to refuse to do any of the work!

Opus 5.5 is by far much better.

Used up 76% of my weekly (5x plan) usage, though.

As of the latest update, I cannot even get Opus 5.5 to create a fictional user in my test database to test the login UI because it refuses to handle password inputs altogether.

It started from one prompt I typed into Claude Code: "*using only pixels directly remake a beautiful pokemon red remake … absolute pixel art perfection.

* No image files at all. The game draws into a 320×180 screen pixel by pixel.

I gave Opus 5.5 my Minecraft mod file and told it to create a Trailer