AI model evaluation platform Arena announced that DeepSEEK-v4.1-Flash-20261017 has ranked first in Agent Arena's open source model and 11th in the overall ranking. Based on approximately 7,500 real-use sessions, the model's overall net improvement over the average model was approximately 5.1%, and the median cost per task was approximately $0.20. This version ranked first in the open source model in the four indicators of successful task confirmation, positive and negative user feedback, and bash command execution reliability, and leadability. Among them, the task success confirmation index ranked 8th in the overall ranking.

AI model evaluation platform Arena announced that DeepSEEK-v4.1-Flash-20261017 has ranked first in Agent Arena's open source model and 11th in the overall ranking. Based on approximately 7,500 real-use sessions, the model's overall net improvement over the average model was approximately 5.1%, and the median cost per task was approximately $0.20. This version ranked first in the open source model in the four indicators of successful task confirmation, positive and negative user feedback, and bash command execution reliability, and leadability. Among them, the task success confirmation index ranked 8th in the overall ranking.