My benchmark is ripping tailwind out of a few year old elixir Phoenix liveview app, and replacing it with component level scoped styles
It's a good and complex task, that requires touching the build system, most components, the stylesheets, and more. Opus 4.6 could barely do it. Sonnet 4 cannot (haven't tried 5 yet). MiniMax actually did fairly well
Grok aced it, rather quickly and cheaply, surprisingly
I run each through Oh my pi, with dexter providing the LSP for elixir
It's a good and complex task, that requires touching the build system, most components, the stylesheets, and more. Opus 4.6 could barely do it. Sonnet 4 cannot (haven't tried 5 yet). MiniMax actually did fairly well
Grok aced it, rather quickly and cheaply, surprisingly
I run each through Oh my pi, with dexter providing the LSP for elixir