Arena launches Agent Arena Index from ~3M real agentic sessions, with six new performance indexes
infwinston · x · 2026-11-19
Arena (LMArena) introduced the Agent Arena Index, its headline measure of overall agent performance built on nearly 3 million real-world agentic sessions, plus six focused indexes: Agentic Work, Coding, Chat, Task Outcomes, Tool Use, and Steering — alongside the existing Alignment Index.
Scores were updated for stability and interpretability. Per the accompanying leaderboard, Claude Fable 5.2 (Max) leads at 1,453 (117 points above rank 2), with Gemini 4 Neon (High) at 1,336; a Pareto frontier chart also compares per-task cost and time across models.
Related event: Arena Launches Agent Arena Index Built on 2.9M Real Sessions(7 posts)→
More from Models
- Claude solves two long-open problems in black hole physics, Anthropic says — burny_tech · 2026-11-19
- GPT-6.2 models keep reusing the same 1992 Numerical Recipes random-number code for grain — every · 2026-11-19
- DeepSeek V4.1 Flash is the best open-weight model on KradleBench's economy dimension — dansitu · 2026-11-19
- Gemini 4 variants take three of top four spots on BrushArena image arena — teortaxesTex · 2026-11-19
- Astra's proof sidesteps three deep Mathlib results by inventing a new 'capacity' quantity — IgorCarron · 2026-11-19
- Ex-Claude superfan: getting the best from Astra/Sol requires relearning how you prompt — mariofilhoml · 2026-11-19