Qwen3.8 Max scores 1739 Elo on GDPval-AA, ahead of Kimi K3 (1685), effectively…

Alibaba's Qwen3.8 Max achieved a stronger professional-task benchmark result while taking many more interaction rounds. Artificial Analysis' August 6 report gives it 1,739 Elo on GDPval-AA, up 468 points from Qwen3.7 Max, with average turns increasing from 14 to 64.

GDPval-AA evaluates completed work products, such as documents and spreadsheets, and uses head-to-head comparisons to produce relative ratings. The improvement therefore concerns the quality of submitted work in that test, accompanied by more interaction rounds per task. The associated cost breakdown explains why this matters when comparing model prices.