Alibaba's Qwen3.8 Max scores 56 on the Artificial Analysis Intelligence Index, a 10-point jump over Qwen3.7 Max. The model catches up to Claude Opus 4.8, but Kimi K3 still achieves a higher score at a 25% lower cost.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
Alibaba's latest Qwen3.8 Max matches Claude Opus 4.8 on the Artificial Analysis Intelligence Index, but Kimi K3 still leads with a 25% cost advantage. However, Qwen3.8 Max's inference cost has doubled, and it shows regressions such as increased hallucination rates.
According to Artificial Analysis data, Qwen3.8 Max scores 56 on the Intelligence Index, up 10 points from its predecessor Qwen3.7 Max's 46, tying with Claude Opus 4.8 but trailing Kimi K3's 57.
Kimi K3's cost per task is $0.86, about 25% lower than Qwen3.8 Max's $1.14, and it scores higher, showing stronger cost-effectiveness.
On the GDPval-AA benchmark, Qwen3.8 Max scores 1739, surpassing Kimi K3's 1685 and trailing only Claude Opus 5's 1852.
However, this high score comes at the cost of a longer reasoning chain: Qwen3.8 Max requires 64 steps per task, while Kimi K3 only needs 14; input tokens increase 15-fold because the test resends the full conversation history at each step.
This leads to a significant rise in inference cost: Qwen3.8 Max's cost per task is $1.14, more than double Qwen3.7 Max's $0.53, despite lower token prices (input drops from $2.50 to $2.00 per million tokens, output from $7.50 to $6.00, and cache hits from $0.50 to $0.25).
Compared to its predecessor, Qwen3.8 Max drops 2 points on AA-LCR, which tests the model's ability to extract information from long texts; AA-Omniscience drops 10 points, which measures the model's accuracy in answering knowledge questions or honestly admitting ignorance.
Accuracy remains around 31%, but hallucination rate jumps from 23% to 40%, indicating the model guesses more often rather than admitting it doesn't know.
The information in this article is primarily sourced from Artificial Analysis benchmark tests, as reported by THE DECODER. All data are as claimed by the source and have not been independently verified.
Qwen3.8 Max matches Claude Opus 4.8 on the Intelligence Index, but Kimi K3 offers a higher score at a lower cost. However, Qwen3.8 Max's inference cost has doubled, and it shows regressions such as increased hallucination rates, indicating that performance gains come with trade-offs.
Primary report
Primary source