Qwen3.8 Max matches Claude Opus 4.8 but trails Kimi K3 at higher cost

What happened

Alibaba's Qwen3.8 Max scored 56 on the Artificial Analysis Intelligence Index, a 10-point jump from Qwen3.7 Max (46), putting it on par with Claude Opus 4.8 but behind Kimi K3 (57). On the GDPval-AA benchmark for work-related tasks, Qwen3.8 Max reached 1,739 Elo points, surpassing Kimi K3 (1,685) but trailing Claude Opus 5 (1,852).

Why it matters

Despite lower token prices, Qwen3.8 Max costs more than double per task ($1.14 vs. $0.53 for the prior version) because it needs 64 steps instead of 14 and resends full conversation history at each step, making input tokens grow 15×. Kimi K3 delivers one point higher on the Intelligence Index for just $0.86 per task, offering better value. The model also shows regressions: hallucination rates jumped from 23 to 40 percent, and it guesses far more often instead of admitting uncertainty.

What to watch

Token pricing dropped (input from $2.50 to $2.00 per million tokens, output from $7.50 to $6.00 per million tokens, cache hits from $0.50 to $0.25), but the efficiency trade-off means cost-per-task remains a key competitive pressure against Kimi K3 and Claude Opus 5.