Opus 5 피드백 메가스레드
Opus 5 feedback Megathread
핵심 요약
Opus 5 사용자들이 겪는 버그, 성능, 작업 방식 변화 등 다양한 경험과 피드백을 공유하는 통합 스레드입니다.
- 작업 효율성 — 작업 완료 후 계속해서 추가 이슈를 생성하여 세션을 길게 늘어뜨리는 문제 발생함
- 모델 언어 품질 — 의미 없는 약어나 문구를 생성하여 문단이 읽기 어려워지는 현상 보고됨
- 의사소통 방식 — 기술적 능력은 뛰어나지만 의사소통이 서툴고 'um, actually' 식의 태도를 보임
- 피드백 형식 — 버그나 이상 행동 보고 시 구체적인 프롬프트와 로그를 포함한 양식 준수 권장
opus 5 has been out for a little while now, and the subreddit is already filling up with separate posts saying it is amazing, lazy, broken, cheaper, more expensive, forgetful, or somehow all of those at once.
so, let’s collect everything in one place.
use this thread to share your actual experience with opus 5: what works, what does not, and what you have changed in your workflow.
what can you share?
- bugs or strange behaviour
- things it does better or worse than previous models
- coding quality and how many revisions it needs
- context handling, compaction, or forgetfulness
- usage and token consumption
- differences between effort levels
- writing style or communication issues
claude.mdinstructions that helped- skills, hooks, subagents, and orchestration setups
- comparisons with opus 4.8, fable, sonnet, codex, or other models
- positive experiences and things it does especially well
reports like shallow codebase investigation, confident guesses, context drift, or unusual usage are welcome, but please include enough detail for others to understand what happened.
reporting a problem?
copy this format:
claude code version:
plan:
effort level:
task:
what you expected:
what actually happened:
can you reproduce it?:
comparison model, if relevant:
screenshots, prompts, logs, and reproducible examples are much more helpful than simply saying “opus 5 is lazy” or “opus 5 is nerfed.”
please redact private code, api keys, company information, and personal details.
a few simple rules
- keep each top-level comment focused on one issue or observation
- check whether someone has already reported the same thing
