TACIT Benchmark

TACIT Benchmark v0.1.0 Transformation-Aware Capturing of Implicit Thought A programmatic visual reasoning benchmark for evaluating generative and discriminative capabilities of multimodal models across 10 tasks and 6 reasoning domains. Author: Daniel Nobrega Medeiros arXiv paper GitHub Overview TACIT presents visual puzzles that require genuine spatial, logical, and structural reasoning — not pattern matching on text. Each puzzle is generated programmatically… See the full description on the dataset page:

Source: Hugging Face Hub (tylerxdurden/TACIT-benchmark). Metadata imported from the dataset’s Hub tags.