Compare Qwen3.8 Max with other leading models at:

This post is the closing link in Artificial Analysis's August 6 thread on Alibaba's Qwen3.8 Max. The thread opened with a score of 56 on the firm's Intelligence Index at $1.14 per task, one point behind the open-weights leader Kimi K3. It found gains over Qwen3.7 Max across agentic, scientific-reasoning and coding evaluations, and a regression on AA-Omniscience, where the hallucination rate rose from 23% to 40%, as detailed in the thread.

The link no longer shows that model. The page at this address now covers Qwen3.8 Max (0902), a later release dated September 2, 2026, which scores 45 on version 4.3.2 of the Intelligence Index. The August model has moved to its own page, marked deprecated, where it now shows 40.

Because the same August release scored 56 in the thread and 40 on today's page, the index has evidently changed since the thread was posted. The thread's figures can be compared with one another, and the live page's figures with one another, but a number from one should not be set against a number from the other.