Qwen3.8-Max matches Claude Opus 4.8 but costs double per task

Alibaba's Qwen3.8-Max scores 56 on the Artificial Analysis Intelligence Index, a 10-point jump over Qwen3.7 Max and level with Claude Opus 4.8, but behind Kimi K3 at 57, which runs 25% cheaper. On GDPval-AA it gained 468 Elo to 1,739, passing Kimi K3 and trailing only Claude Opus 5 at 1,852. It gets there by taking 64 steps per task instead of 14, with input tokens up 15x. A single Intelligence Index task now costs $1.14, more than double Qwen3.7 Max's $0.53, despite lower per-token prices. An independent VulcanBench run put the model mid-pack at best effort and last at default, because Alibaba's own coding benchmarks allowed a five-hour timeout against VulcanBench's 45 to 60 minutes. The lesson: cost per successful task, with explicit time budgets, is the only figure that predicts the bill.

View full digest for August 7, 2026