Qwen3.6-35B-A3B on my laptop drew me a better pelican than Claude Opus 4.7
Summary
Simon Willison compares Qwen3.6-35B-A3B and Claude Opus 4.7 using his 'pelican riding a bicycle' benchmark, running Qwen on a MacBook Pro M5 via LM Studio with a 20.9GB quantized model. The post notes that Opus 4.7 performs worse on this task and emphasizes the benchmark's humorous nature while acknowledging its usefulness as a rough proxy for model practicality. It highlights considerations around local LLM deployments, model size, and tangible tasks like SVG illustration for small teams evaluating AI tools.