Qwen3.6 Triumphs Over Opus 4.7 in Humorous Benchmark Test

In a playful yet insightful comparison, I tested two newly released AI models—Qwen3.6-35B-A3B from Alibaba and Claude Opus 4.7 from Anthropic—using my unconventional 'pelican riding a bicycle' benchmark. The results were amusing: Qwen3.6 outperformed Opus, which struggled with the bicycle frame. Even when I adjusted Opus to its maximum thinking level, it didn't improve significantly. To further test these models, I asked them to generate an SVG of a flamingo on a unicycle. Once again, Qwen impressed, even adding a witty touch with a sunglasses comment on the flamingo.

While the pelican benchmark started as a joke, it surprisingly correlated with the models' general utility over time. Early attempts were poor, but recent models like Gemini 3.1 Pro have produced usable illustrations. However, today's results suggest that this correlation may no longer hold. Despite my respect for Qwen, it's hard to believe a 21GB quantized model surpasses Anthropic's latest release in overall utility.

Yet, if you specifically need an SVG of a pelican on a bicycle, Qwen3.6-35B-A3B is currently your best choice, even running on a laptop. This quirky benchmark highlights the challenges and humor in evaluating AI models, reminding us of the absurdity in comparing their capabilities.