FrontierHarness: Same Model, 17-Point Gap Across Agent Harnesses
FrontierHarness, a new benchmark from the Guanlan and Legora teams, tested 12 coding agent harnesses over 360 runs and found the same model's score can vary by 17 points depending on the harness, with harness costs differing up to 17x. Codex topped the leaderboard at 66.7%.
2026-09-03 ~ 2026-09-03 · 2 related posts
- FrontierHarness eval: 12 agent harnesses, same model — pass rates span 50%–67%, cost per pass $1.05–$18.34 — stuffyokodraws · 2026-09-03
- Legora's FrontierHarness eval: Codex tops at 66.7%, harness cost spreads 17x — chetanp · 2026-09-03