Agent Arena Launches Agent Reliability Leaderboard

Arena launched the Agent Arena leaderboard based on over 2.38 million sessions and 53 models, measuring agent reliability via user-confirmed task success. Open-source models have matched Claude within three months at 89-96% lower cost, with Claude Opus 5.5 leading in net improvement.

2026-10-13 ~ 2026-10-13 · 2 related posts

• Agent Arena: open models match Claude Fable 5 in 12 weeks at up to 96% lower cost — arena · 2026-10-13
• Agent Arena leaderboard: Claude Opus 5.5 tops net improvement across 2.38M sessions — arena · 2026-10-13