Is Opus 5.5 entering a “nerfed” phase? LiveNerf baseline update
An open-source project has been released to independently measure changes in Opus 5.5's performance following its release. This tool tracks daily performance using established benchmarks like SWE-bench and SciCode, employing a standardized scientific methodology to ensure measurable and reproducible changes over time. The project aims to determine if Opus 5.5 is entering a "nerfed" phase, allowing the community to monitor its performance evolution.
Time & source
Times shown in UTC
Display time zone: UTC
Local time zone unavailable; showing UTC.
PublishedOffset at this time: UTC+0Sep 29, 2026, 20:51 UTC
IngestedOffset at this time: UTC+0Sep 30, 2026, 00:00 UTC
- Published
- Sep 29, 2026, 20:51
- Ingested
- Sep 30, 2026, 00:00
- Source type
- Dev community
- Tier
- Community
- Source status
- Healthy
Tier is a per-source editorial setting, not a per-item score.
A few days ago, I released an open-source project that gives the Claude community a way to independently measure whether Opus 5.5’s performance changes or gets “nerfed” in the weeks following its release. It tracks performance day by day using established benchmarks such as SWE-bench and SciCode, applying a standardized scientific methodology to make changes over time measurable and reproducible. If you’re interested in tracking the performance of Opus 5.5, here’s the repo:
We are still in the evaluation phase, so I wanted to clarify since I saw some people asking if today’s dip means there’s for sure a nerf occurring right now, we won’t have that data until day 20 after the model release. So we’re still about two weeks out from having data to support that one way or another. Today’s dip will be factored into the baseline which will happen at day 10.
I also want to encourage anyone to look at how I’m collecting this data and bring forward any improvements you’d suggest for future benchmarks, since I do plan on running more in the future, however I’m self funding this and calibration costs a good amount. It took roughly one week’s worth of a pro plan to do Opus 5.5.
I strongly believe in transparency, AI companies are not being transparent, so we will make our own tools that reveal things currently hidden.
A huge thanks to everyone who’s already supported LiveNerf, I hope to not let you guys down. It’s gotten bigger than I imagined it would, I was just going to run this scientifically for myself, but apparently the community has enjoyed it. Very grateful to you all. If we do detect a potential nerf this month, I’ll be sure to make a post announcing that data.