Livenerf tracks whether Claude Opus 5.5 quietly gets worse after launch

MVMaker Views #AI #Hacker-News

Livenerf tracks whether Claude Opus 5.5 quietly gets worse after launch

The method section shows how to test "it got worse" claims instead of trading anecdotes.

Livenerf is a public benchmark started on Claude Opus 5.5’s launch day that runs frozen prompts daily and measures drift statistically, to test recurring claims that models are "nerfed" after release. It is built on the UK AI Security Institute’s Inspect framework.

Read it at the source

GitHub (ninjahawk/livenerf)livenerf: benchmark for tracking model capability after release ↗ Hacker NewsDiscussion: Livenerf: Has Opus 5.5 been nerfed yet? ↗

Cascadic Analysis 1

UndertowWe don't completely understand why frontier models behave as they do

What hidden risk could pull against this, even if the news is good?

Models can change behind the same name, and users can’t look inside to check. Independent, repeated measurement is one of the few ways to know, because a launch-day review can’t vouch for next month.

Livenerf tracks whether Claude Opus 5.5 quietly gets worse after launch

Further down the stream