Qwen3.8 Max Reclaims Top Spot on China's AI Leaderboard With Benchmark Score of 45
The updated Qwen3.8 Max (0902) scored 45 on the Artificial Analysis Intelligence Index, gaining 5 points in a single month. This update allows the model to surpass GLM-5.3 (44.9) and Kimi K3 (43.8) to take back the top spot on China's AI model leaderboard. This development underscores the fierce competition and rapid iteration speed among Chinese frontier model developers. It demonstrates how continuous post-training refinements on massive Mixture-of-Experts architectures can yield noticeable benchmark gains within short timeframes. The 5-point performance jump comes from a 30-day upgrade applied to the 2.4-trillion parameter Mixture-of-Experts (MoE) model. The narrow 0.1-point margin over GLM-5.3 highlights how tightly contested top-tier model benchmarks have become.
## BACKGROUND
The Artificial Analysis Intelligence Index is a composite evaluation benchmark that combines multiple tests across mathematics, science, coding, and reasoning to measure overall model capability. Mixture-of-Experts (MoE) is an architecture that routes inputs to specialized sub-networks, enabling models to scale up parameter counts dramatically without a proportional increase in computational cost per token.