Qwen3.6-35B-A3B on my laptop drew me a better pelican than Claude Opus 4.7

Article

• Simon Willison ran Qwen3.6-35B-A3B locally; beat Opus 4.7 on SVG pelican drawing
• Model runs on 128GB M5 MacBook Pro via MLX
• Creative/visual gap to frontier is closing on Apple Silicon

Discussion

• Coding benchmark commenter showed Qwen 3.6 solved only 11/98 tasks vs Opus 95/98 — narrow creative win doesn’t generalize
• Commenters noted Opus/Sonnet have been regressing on non-coding tasks since 4.1
• Others praised Qwen 3.5 35B agentic tool-call quality for local use
• Skeptics called pelican benchmark stale and easy to overfit

Discuss on HN