BullshitBench ranking: Claude Opus 4.8 tops at 95%, new Gemini 4 models score surprisingly well

burny_tech · x · 2026-11-19

A BullshitBench v2 leaderboard (266 models) shared by petergostev ranks models on rejecting false premises: Claude Opus 4.8 (no reasoning) and Qwen3.8 Max lead at 95% Clear, followed by Claude Opus 4.8 xHigh (94%) and Claude Sonnet 4.6 (91%). The new Gemini 4 Hydrogen scores 90% (#6) and Gemini 4 Neon 84% (#11), prompting the comment that Google models are doing "super well." Some Claude variants show notable refusals, e.g. Claude Fable 5 at 33%.

Original post →

More from Models

Models channel →