BREAKING
Agent Arena Launches Causal Tracing
0
sessions
0
models
Chatbot Arena vs Agent Arena
Chatbot Arena
Old
●
Human preference voting
●
Elo-style ranking
Agent Arena
New
●
Causal tracing of traces
●
Objective signals
Five Evaluation Signals
1
Confirmed success
↓
2
Praise vs complaint
↓
3
Steerability
↓
4
Bash recovery
↓
5
Tool hallucination
Net Improvement Leaders
Claude Fable 5
14.17
GLM 5.2 Max
9.4
Live Leaderboard Updates
AI NEWS BLITZ
Arena.ai has launched a new leaderboard to evaluate AI agents using causal tracing.