Theo pins the Opus 5.5 nerf reports on the reasoning summarizer
A per-account change to how reasoning traces are rendered, not a quieter model, is Theo's read on the November 3 shift.
Theo looked into the claims that Opus 5.5 had been quietly nerfed and came back with a different explanation. Anthropic appears to be rolling out a change to the reasoning summarizer on a per-account basis, he wrote, adding that he sees no sign of degraded capability himself. His guess is that the company is trying to fix the reasoning traces, which he says have been bad for weeks and which he had complained about the day before.
No sign of degraded capability for me.
What set it off
The post he was responding to came from Shane Melaugh, who asked whether Opus 5.5 was nerfed and said he had Gemini 4 Argon analyze 1,400 of his Claude Code sessions. The finding, as he describes it, is that something definitely happened on November 3. The model name did not change, but the way the model does reasoning took a sharp turn.
the model name didn't change, but the way it does reasoning took a sharp turn.
That is a real signal in a real corpus, and it is worth separating from the conclusion people reach for. A change in the shape of reasoning output is not the same measurement as a change in the quality of code the model produces. Melaugh's analysis is of sessions, not of task outcomes, and Theo's counter is a single person's experience, which is weaker evidence still. Nobody in this exchange has posted a before and after on a task suite.
Why the distinction matters
The reasoning trace you see in a client is a rendered summary of the model's thinking, produced by a separate summarizing step. If that step changes, every trace in your terminal looks different on the same day, across accounts, with no model version bump and no release note. That is exactly the fingerprint people read as a stealth downgrade, because it is the only part of the system most users can watch directly.
So the honest state of things is unresolved. There is a dated, corpus-wide observation that reasoning output shifted, and there is one well-known user saying his results did not. Neither has been tied to an announcement from Anthropic, and nothing in the exchange confirms what the change was or who got it.
Theo's own follow-up a few hours later was less analytical. "Going back to Opus 4.5 just to feel something," he posted, which reads as a joke rather than a recommendation, and should be filed accordingly.
Have the good opinion, not the loud one
One email a day. Under three minutes. What working engineers found out about coding with AI.