Kimi K3 (thinking) vs Qwen3.7 Max (default)

Kimi K3 in thinking mode at #12 against Qwen3.7 Max at #50. 89.25 overall to 81.73, so eight points, and 2416.4 Elo against 1699.1. The single most lopsided metric is length adherence, and it's Kimi's best trick: 91.10 against 87.67. Nearly seven points, and 91.10 is the strongest length control of any model in these comparisons. On a long brief that's the difference between a draft you trim and a draft you restructure. Continuity is the other clear win, 88.28 to 79.03. Qwen's hooks at 84.25 are its most competitive number, about six back. Price runs the other way: Qwen is about five cents against Kimi's 21, so four times cheaper, and it's twice as fast. But Qwen is closed weights, so the cheap-and-open argument that makes GLM-5 or MiniMax interesting doesn't apply here. You're paying less for less, without the hosting upside.

Pick Kimi K3 (thinking) if length discipline and flow across a long script matter, and you want open weights.
Pick Qwen3.7 Max if you're on Alibaba infrastructure and want four times cheaper drafts with solid openings.