Bongochat Tops GPQA-Dumb Benchmark: Zero Coding, All Math Wrong

TheOnlyVibemaster · reddit · 2026-08-04

A developer has released Bongochat, an open-weight spoof model that officially ranks first globally on the GPQA-Dumb benchmark (where lower scores are weighted higher).

The model's performance is purely disastrous: it blanks on math questions by always picking 'A', scores 0/500 on coding tests, and completely lacks conversational memory. The creator joked that using it feels "pretty much like using Gemini or Grok."

Related event: Bongochat tops the dumbest AI model leaderboard(2 posts)→

Original post →

More from Fun

Fun channel →