FeeligWang @FeeligWang - Twitter Profile

Pinned Tweet

about 16 hours ago

Glad to meet you on X Keep sharing what I have found interesting and practical on AI and tools Why not make friends? F4F all the time

4

11

0

207

FeeligWang

@FeeligWang

6 minutes ago

https://t.co/YJEUrj1XFy

0

7

FeeligWang

@FeeligWang

6 minutes ago

Don't jump Claude models when a task stalls. Addy Osmani's official guide, via @ClaudeDevs: start most work on Opus 5.5 at default medium - half Fable 5.2's per-token price. Change effort first. Past high, one more effort level can cost as much as a bigger model. Haiku 5.5: classify, route, file-search and log subagents. Sonnet 5.5: scoped bugs, specced features, repeated agent work. Opus 5.5: everyday multi-PR engineering and computer use. Fable 5.2: major refactors, migrations, architecture, leading subagents, deep research. If it falls short: add a self-check (test, build, or script); raise effort one level; only then move up a model, from lower effort. Official SWE-bench Pro subset: Opus 5.5 at low, failures rerun at high, about 97% / $0.17 vs 95.3% / $0.29 all-high. Not a rerun. Mix at a break (/compact or a short plan). Keep edits on the main model; subagents on Haiku or Sonnet. Defect-search bench: Opus lead + up to 25 Sonnet workers cost about a third less than