Paper reveals Agent benchmark unreliability: Harness variance 7.8x model variance
omarsar0 · x · 2026-08-26
A new paper investigates why agent leaderboard comparisons are hard to trust, focusing on how much of a benchmark score is attributed to the "harness"—the layer between the model and the task.
The harness builds context, mediates tool calls, validates outputs, and decides when to stop. In a controlled experiment on 100 tasks from SWE-bench Verified, researchers tested three frontier models against three harness configurations while holding other variables constant.
Results showed that swapping the harness caused GLM-5.1's score to swing by 13.0 points, whereas swapping the model within a fixed harness caused changes of only 3.0, 2.5, and 5.0 points. Harness-induced variance was found to be 7.8x larger than model-induced variance, with 6 out of 9 model-pair rankings flipping depending on the harness used.
Related event: Paper finds agent benchmarks unreliable as harness matters more than models(2 posts)→
More from coding & agent
- AI software factories hinge on the Factory Loop, not faster code generation — Pavan_Belagatti · 2026-10-16
- 'Dropping a 3,000-line PR twenty minutes after ticket assignment' — AI coding meme goes viral — facontidavide · 2026-10-16
- Self Labs launches Agent Vault to let AI agents log in and pay with credit cards — EXM7777 · 2026-10-16
- Plane Agents: triggers, not prompts, are what turn chatbots into workflow agents — JosephJacks_ · 2026-10-16
- Claude eng lead Felix Rieseberg publishes tutorial on being more ambitious with AI — danshipper · 2026-10-16
- Qwen Code Desktop v0.25.2 ships managed-agent fixes and MCP config repair — github-actions[bot] · 2026-10-16