Arena Launches Agent Arena Index Built on 2.9M Real Sessions
Arena (the team formerly known as LMArena) released the new Agent Arena Index on November 19, replacing the previous Agent leaderboard, along with an Alignment Index and six specialized sub-leaderboards. The new index system is built on 2.9M+ real user sessions and millions of agent task trajectories from a global community, measuring agents' overall ability to complete complex workflows, use tools, and adapt to feedback. A methodology article was also published, and the index is available to browse online.
Confirmed
- The Agent Arena Index is based on 2.9M+ real user sessions and provides a comprehensive score across all agent task types, replacing the previous Agent leaderboard.
- Beyond the Agent Arena Index and Alignment Index, six focused leaderboards were added, each answering a specific model-selection question, covering scenarios such as Work (knowledge work, combining four directions), coding, office tasks, and chat.
- The data comes from real tasks and millions of agent sessions/trajectories.
- The methodology article has been published, and the Agent Arena Index can be viewed online.
Why it matters
- The rankings move from a single overall score to scenario-specific indexes, directly serving the decision of "which model for which task."
- Grounded in real user sessions rather than constructed benchmarks, it reflects how agents perform in real-world workflows.
Additionally, the post's author relayed that Claude leads three of the sub-leaderboards ("sweeping three boards"), a claim from community reposts—official online rankings are the authoritative source for exact standings.
2026-11-19 ~ 2026-11-19 · 8 related posts
Primary sources
- LMArena overhauls agent leaderboard: Agent Arena Index scores 2.9M sessions with anchored 1000-point scale — arena ·
- Arena launches Agent Arena Index and Alignment Index built on millions of real tasks — arena ·
- Arena launches Agent Arena Index from ~3M real agentic sessions, with six new performance indexes — infwinston ·
- [source] LMArena overhauls agent leaderboard: Agent Arena Index scores 2.9M sessions with anchored 1000-point scale — arena · 2026-11-19
- LMArena launches six new agent indexes; Claude Fable 5.2 tops Work, Coding and Chat — arena · 2026-11-19
- Arena launches six new agent indexes from millions of real sessions, updates score scaling — arena · 2026-11-19
- Arena points readers to the live Agent Arena Index and its methodology — arena · 2026-11-19
- [source] Arena launches Agent Arena Index and Alignment Index built on millions of real tasks — arena · 2026-11-19
- [source] Arena launches Agent Arena Index from ~3M real agentic sessions, with six new performance indexes — infwinston · 2026-11-19
- Agent Arena Index launches: ~3M real-world agentic sessions benchmark agent performance — infwinston · 2026-11-19
1 near-duplicate retellings: infwinston