The DeepSeek-V4.1-Flash-1017 model has reached the top of the Agent Arena and Code Arena: WebDev open-source leaderboards. According to the platform, the model holds the 11th position in the overall rankings.
The evaluation is based on 7,498 real-world sessions, showing a net improvement of 5.13%. The model leads among open-source entries in task success confirmation at 11.54%, user feedback at 5.89%, and Bash command reliability at 5.42%.
The model recorded a tool hallucination rate of 0.48%, placing it 7th in that specific metric. Its steerability index stands at 2.46%, which also ranks first among open-source models.
Regarding operational expenses, the median cost per task for the model is between $0.18 and $0.19. The model is currently included in the Pareto frontier for performance and cost efficiency.
