Qwen 3.6 Small MoE Model Running Locally Outperforms Claude Opus 4.7 on SVG Generation Benchmark

Summary

Simon Willison, a prominent developer and AI commentator, reported that Alibaba's Qwen3.6-35B-A3B mixture-of-experts model — running locally on his laptop — produced a superior SVG rendering of a pelican riding a bicycle compared to Anthropic's Claude Opus 4.7. The Qwen model uses only 3 billion active parameters despite having 35 billion total parameters, making it lightweight enough for consumer hardware inference. While Willison himself notes his 'pelican riding a bicycle' SVG generation tes