Show HN: FrontierHarness Eval – 9 harness, same model, cost per pass varies 17x
Summary
The article presents a comprehensive benchmark of FrontierHarness Eval, comparing 9 harnesses across 12 configurations on identical software engineering tasks using the same model and runtime. It reports pass rates, median cost per task, cache hit rates, and median runtimes, while noting methodological safeguards and caveats such as failures excluded for some tasks and potential warm-cache bias. It also provides guidance on running your own harness tests on Runta and lists the tested harness versions.