An open-source monitoring project called LiveNerf says it has tracked Claude Opus 5.5 for 38 days and has detected no performance degradation so far, as users continue to question whether the model has quietly gotten worse.
Day 38 · LiveNerf tracker
No degradation detected in Claude Opus 5.5 so far, but the tracker can only see changes of 7% or more
An open-source, community-run project reports 38 days of monitoring and no evidence of quality decline. Its current sensitivity is limited, and its stated goal is to push detection below 1%.
38
days of data collected on Opus 5.5
0
degradations detected so far
7%
current detection floor (smallest change it can flag)
<1%
stated target detection floor
Smallest change the tracker can detect
Each block represents 1% of change. Drawn to scale.
Reaching the target means seeing changes roughly seven times smaller than the current limit.
What “no degradation” does and does not mean
A clean result at a 7% floor does not rule out smaller shifts
Changes below 7% fall outside what the tracker can currently flag.
The specific tests, sample sizes and validation methods are not described in the update.
Findings are the project’s own reporting and have not been independently audited.
Why it matters
Developers and businesses building on hosted models need consistent behavior. A public, community-run record of model performance over time offers a way to check claims that are otherwise hard to verify from outside.
Open questions
It is unclear whether this approach can reach the sensitivity needed to catch subtle regressions. Progress will depend on narrowing the threshold from 7% toward 1%.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...