New DeepSeek V4.1 Flash Version Tops Agent Arena Open-Source Model Leaderboard, Ranks 11th on Overall Leaderboard

The AI model evaluation platform Arena announced that DeepSeek-V4.1-Flash-20261017 (Max) has secured the top spot among open-source models on Agent Arena, and ranks 11th on the overall leaderboard. Based on approximately 7,500 real-world usage sessions, this model delivers an approximate 5.1% overall net improvement over the average model, with a median per-task cost of around $0.20. This version ranks first among open-source models across four metrics: task success confirmation, positive and negative user feedback, Bash command execution reliability, and steerability, with the task success confirmation metric ranking 8th on the overall leaderboard.

Source: PANewsRead original