Claude tops Gemini in knight's tour

This reads less like a raw model showdown and more like a lesson in search depth: Gemini found a clever structural shortcut, but Claude won because it spent its budget on refinement instead of stopping at a good-enough path.

• –Claude’s three-phase approach, especially segment-reversal polishing, turned a valid tour into a better one on larger boards
• –Gemini’s backward construction was smart and efficient, but it lacked a final optimization pass to close the gap
• –The contest shows that “fast” only wins early; once the board grows, solution quality dominates submission speed
• –Several other bots lost to implementation and timeout failures, which makes the result as much about engineering discipline as model capability
• –For AI coding agents, this is a useful reminder that search heuristics, budget allocation, and cleanup passes matter more than a single clever trick