데일리 드라이버로 Sonnet 4.6을 쓰는 중인데, 제가 뭘 놓치고 있는 걸까요?
Sonnet 4.6 as a daily driver. Am I missing something?
핵심 요약
Opus의 속도와 제한 문제로 Sonnet을 구현용으로 사용했더니 훨씬 효율적이라는 경험 공유.
- 워크플로우 최적화 — Opus로 계획을 세우고 Sonnet으로 구현하는 방식이 훨씬 빠르고 효율적임.
- 성능 및 비용 — Opus는 속도가 느리고 제한에 자주 걸리지만, Sonnet은 빠르고 실수가 적음.
- 명확한 지시사항 — Sonnet을 효과적으로 쓰려면 짧고 명확한 작업 단위와 지시가 필수적임.
- 내장 기능 활용 — Claude Code의 opusplan 별칭을 사용하면 계획과 구현을 자동으로 분리할 수 있음.
Long-time CC user. Got a LOT of mileage out of sonnet and opus 4.5 last year. This year my company has put devs on max plans so started defaulting to Opus for everything. Did this all the way through 4.6 and now 4.7.
With 4.7 I would hit limits pretty quickly whenever I kicked off parallel work. It is also insanely slow. For context I feel i do my best work when I am focused, reviewing model implementations and actively testing. With Opus 4.7 the delays started to break my flow pretty badly.
Recently decided to switch to sonnet as the implementer. Opus analyzes requirements and writes up an implementation playbook (in to-do list format)
Sonnet just implements in a loop. Instructed to commit after every task completion (with tests included), run in a loop and only stop around 300k context window usage.
It is seriously powerful. I can test things as it goes, it moves fast and makes very few mistakes.
I’m curious if I’m leaving something on the table here but frankly this feels like the way Claude code was always meant to work. A 45 min session to implement 3 basic features feels insane when you can do 3x as much in 30 minutes with Sonnet.
