Loop & Harness EngineeringZhang et al. (arXiv:2605.23950)

Stop Comparing LLM Agents Without Disclosing the Harness

Stop Comparing LLM Agents Without Disclosing the Harness

Why it matters

Harness-induced variance was 7.80× model-induced variance, and rankings flipped in 6 of 9 model pairs

Source — Primary sources first; secondary reports are labelled “Press”

Paperarxiv.orghttps://arxiv.org/abs/2605.23950

Added in Week 35, 2026 · See that week

Same week

Iwase et al. publish Reliable Chain-of-Thought via Prefix ConsistencyPaperTeaching Claude WhyOfficialGPT-Realtime-2 / -Translate / -WhisperOfficialMoonshot AI raises $2 billion at a $20 billion valuationPress

Nearby on the same axis

Agent Harness EngineeringOfficialLoop EngineeringOfficialWhat makes a harness a harnessPaperHarness engineering for coding agent usersOfficial