Qwen3.8 Max nearly catches Claude Opus 4.8 — but Kimi K3 still scores higher

Alibaba's Qwen3.8 Max scored 56 on the Artificial Analysis Intelligence Index, a composite benchmark spanning nine evaluations including GDPval-AA, Terminal-Bench, SciCode and Humanity's Last Exam. That's a 10-point jump over its predecessor, Qwen3.7 Max, which scored 46 — and it puts Qwen3.8 Max exactly level with Claude Opus 4.8 in its max configuration.

The score puts Qwen3.8 Max ahead of every model out of Google, Meta and xAI, behind only the top-tier releases from Anthropic and OpenAI and Moonshot AI's Kimi K3, which scores 57. On cost, though, Kimi K3 comes out ahead too: it averages $0.86 per task on the Intelligence Index versus $1.14 for Qwen3.8 Max. Qwen3.8 Max's lower per-token rate ($2 per million input tokens) gets offset by how verbose the model is — it burns through roughly 150 million tokens to complete the benchmark, against a median of 66 million across the index. The model itself launched on August 3, 2026: a mixture-of-experts design with 2.4 trillion total parameters and roughly 95 billion active per step, with a lighter open-weight Qwen3.8-27B variant following the same week.

• Qwen3.8 Max: 56 points, up 10 from Qwen3.7 Max's 46
• Kimi K3: 57 points — a point above both Qwen3.8 Max and Claude Opus 4.8
• Kimi K3 costs about 25% less per task ($0.86 vs $1.14 for Qwen3.8 Max)

Both Chinese models are part of the same open-weight release wave that started with Kimi K3's full weights going public in July. Anthropic is answering with its own price play: Claude Opus 5 offers near-frontier intelligence at half its predecessor's price — the race is no longer just about benchmark scores, it's about what each completed task actually costs.

None of this should be read as personalized investment advice.