POV — Everyone Read The Wrong Number (Claude Opus 5)

Anthropic shipped Claude Opus 5 — within 0.5% of the flagship on CursorBench at half the cost, and it BEAT the flagship on OSWorld at a third the cost. Consensus says the flagship is dead. I think that's the wrong read, and 'benchmarks are meaningless' is lazier still. Benchmarks are qualifying laps; production agent work is Le Mans — and the pole-sitter usually doesn't win Le Mans. What separates the tiers is endurance, and no leaderboard measures it. Watch: https://youtube.com/watch?v=NHBf2x5-rhk