Releases·Tool Releasefeatured
LiveNerf project tests whether Anthropic nerfs Opus 5.5 over time
A developer built LiveNerf, an open-source benchmark that re-runs fixed tests on Claude Opus 5.5 daily to independently detect post-release capability degradation. The project reached ~1,000 GitHub stars and the Hacker News front page, and the 10-day baseline is now complete, with day-30 data collection beginning. No nerf has been detected yet; meaningful judgment is deferred until day 20.
5527 engagement6 sourcesTue, Sep 22, 2026, 11:15 PM
Developments
10-day baseline completed on 2026-10-03; day-30 comparison data collection now underway.
Sat, Oct 3, 2026, 05:19 PM
Day-9 update: a recent score dip will be factored into the day-10 baseline; nerf verdict data not available until day 20.
Tue, Sep 29, 2026, 08:51 PM
AI Overview
- Redditor ninjahawk launched a month-long daily benchmark testing whether Anthropic nerfs Opus 5.5 over time, logging results at github.com/ninjahawk/livenerf
- The benchmark uses fixed deterministic tests designed with Claude so capability regressions later in the month would show; community PRs welcome after review
- The tester suspects Anthropic may apply exponential quantization over time and says the month-long run will independently verify or refute it
- No nerf detected so far; a deviation above 7.5 points versus an Opus 5 control benchmark would flag potential capability loss
- No verdict yet; ninjahawk won't interpret daily movements until the baseline-to-day-30 comparison, with no meaningful degradation also counting as a result
- The benchmark tracks Opus 5.5 daily on established benchmarks including SWE-bench and SciCode
- ninjahawk is self-funding the run; calibrating Opus 5.5 took roughly one week of a pro plan
- Day 8: LiveNerf repo nears 1,000 stars and hit the Hacker News front page; baseline still being established before degradation testing begins
Related
Social·Community
LiveNerf Tracks Claude Opus 5.5 Daily to Test Post-Release 'Nerfing', Establishes 58.4% Baseline
A developer built LiveNerf, an open-source benchmark that measures Claude Opus 5.5's performance day by day to test whether models degrade after release. Ten days of measurements established a baseline of 58.4%, and by day 20 the project expects data on whether post-release changes occur.
A developer built LiveNerf, an open-source benchmark that measures Claude Opus 5.5's performance day by day to test whether models degrade after release. Ten days of measurements established a baseline of 58.4%, and by day 20 the project expects data on whether post-release changes occur.
Updated Tue, Oct 6, 2026, 07:28 PM — Baseline established at 58.4% over days 1-10; measurable nerf-check period underway, results expected by day 20.
547 engagement1 sourceredditTue, Oct 6, 2026, 07:28 PM
Social·Community
Are They Nerfing Opus 5.5? Livenerf Day 20 update
The LiveNerf project, a community effort to detect post-release capability degradation ('nerfs') in Anthropic's Opus 5.5, published its day-20 update. After three weeks of daily measurement, variance decreased after the baseline period, suggesting regression to the mean rather than a nerf. The author plans a final verdict in 10 days and asks the community to contribute anecdotal reports of perceived capability drops.
The LiveNerf project, a community effort to detect post-release capability degradation ('nerfs') in Anthropic's Opus 5.5, published its day-20 update. After three weeks of daily measurement, variance decreased after the baseline period, suggesting regression to the mean rather than a nerf. The author plans a final verdict in 10 days and asks the community to contribute anecdotal reports of perceived capability drops.
684 engagement1 sourceredditWed, Oct 14, 2026, 07:10 PM
Social·Community
Opus 5.5 Context Degradation
A Reddit user describes testing Anthropic's Opus 5.5 through extreme context lengths on a large pipeline build. They report no performance loss or serious hallucination around 500k tokens, with degradation only past ~700k — a large improvement over Opus 5, which reportedly degraded near 300k.
A Reddit user describes testing Anthropic's Opus 5.5 through extreme context lengths on a large pipeline build. They report no performance loss or serious hallucination around 500k tokens, with degradation only past ~700k — a large improvement over Opus 5, which reportedly degraded near 300k.
12 engagement1 sourceredditWed, Sep 23, 2026, 05:09 PM
Social·Community
Sonnet 5.5 vs Opus 5.5 vs Sonnet 5 in Claude Code and pi: results and behavioral differences on 10 coding tasks
Multiple independent developers benchmarked Anthropic's new Sonnet 5.5 against Opus 5.5 and Sonnet 5 in Claude Code and pi-agent. Sonnet 5.5 matched Opus 5.5's pass rates while running roughly twice as fast and costing one-third to one-half as much, and one tester with skills loaded could not distinguish the two models. A separate comparison found Opus 5.5 itself takes far fewer steps and tokens than Opus 5.
Multiple independent developers benchmarked Anthropic's new Sonnet 5.5 against Opus 5.5 and Sonnet 5 in Claude Code and pi-agent. Sonnet 5.5 matched Opus 5.5's pass rates while running roughly twice as fast and costing one-third to one-half as much, and one tester with skills loaded could not distinguish the two models. A separate comparison found Opus 5.5 itself takes far fewer steps and tokens than Opus 5.
Updated Tue, Sep 29, 2026, 04:23 PM — A skills-focused rerun on 2026-09-29 found Sonnet 5.5 indistinguishable from Opus 5.5 across 3 skills and 3 runs each.
11148 engagement3 sourcesredditThu, Sep 24, 2026, 08:15 PM
Releases·Model Releasefeatured
Anthropic launches Claude Opus 5.5 with price reduction
Anthropic released Claude Opus 5.5 via its own launch page on anthropic.com, and the release includes a price reduction relative to prior Opus pricing. The launch drew heavy engagement on Hacker News and Reddit. A later post dated 2026-09-28 links to an anthropic.com page titled 'Claude Sonnet 5.5', suggesting a subsequent Sonnet 5.5 release.
Anthropic released Claude Opus 5.5 via its own launch page on anthropic.com, and the release includes a price reduction relative to prior Opus pricing. The launch drew heavy engagement on Hacker News and Reddit. A later post dated 2026-09-28 links to an anthropic.com page titled 'Claude Sonnet 5.5', suggesting a subsequent Sonnet 5.5 release.
Updated Mon, Sep 28, 2026, 06:18 PM — A Hacker News post dated 2026-09-28 links to anthropic.com for a Claude Sonnet 5.5 release.
1204.4k engagement4 sourcesreddit, hackernewsTue, Sep 22, 2026, 04:29 PM