contradish / flip

Same question.
Different answer.
No warning.

Pick an example. Watch the model flip its conclusion based on minor rephrasing. Not just its tone. Its actual answer.

Choose an example

Most AI systems are tested for correctness. Almost none are tested for consistency. contradish detects conclusion flips automatically, across 3,840 strain tests and 20 domains.

pip install contradish
leaderboard → full demo → how it works →