FrontierCode v1.1 benchmark data reveals that Claude Opus 5 peaks at medium reasoning effort, delivering near-top scores at roughly half the compute cost of higher effort settings.