GeoDuel pits players against open models

GeoDuel has a stronger hook than another model leaderboard: it makes multimodal evaluation playable. Its success will depend on whether model explanations become genuinely auditable rather than just entertaining guesses.

• –The preview shows a player-first flow with image zoom, map pins, and six models answering after the human guess.
• –Arena, Daily Five, and Leaderboard modes suggest a repeatable competitive loop.
• –It taps into active GeoGuessr evaluation work, including GeoBench and research showing that accurate geolocation does not guarantee reliable visual explanations.
• –Transparent prompts, fixed image provenance, and per-model scoring will be essential for meaningful comparisons.