Gemini 4 Argon and Claude Opus 5.5 split across coding tests

The New Stack’s one-run tests favored Opus on billing and long context; Paddo.dev’s 20-run coding test favored Argon on pass count but cost more.