Sorting Agent Arena by 'Praise vs Complaint' aligns rankings with user preferences
MilesCranmer · x · 2026-08-15
A user found that sorting the Agent Arena leaderboard by "Praise vs Complaint" makes the rankings better align with their actual preferences. For instance, this fixes the anomalous scores where Opus ranked higher than Fable. The leaderboard dynamically ranks 48 models based on signals like tool reliability, task completion, and steerability.
More from coding & agent
- LLM Zoomcamp Student Builds Agentic RAG Assistant to Explore Capstone Projects — Al_Grigor · 2026-08-15
- Dynamic System Prompt Assembly Framework: Modular Optimization in Practice — blaizedsouza · 2026-08-15
- AI Mock Tool 'aimock' Hits 1M Weekly Installs, Supporting Full-Stack Simulation — Roger_M_Taylor · 2026-08-15
- 2023 HN thread dismissed Cursor as "crypto bro bullshit" — IndraVahan · 2026-08-15
- Claude Code skill generates semantic chapters and burns bilingual subtitles — tom_doerr · 2026-08-15
- The Next Token Ep. 03: Discussing Agent Uncertainty with Industry Experts — threepointone · 2026-08-15