24 Black-Box LLM Uncertainty Methods Benchmarked Across 4 Models — No Clear Winner

A systematic benchmark of 24 black-box uncertainty estimation methods — spanning verbalization, sampling, explanation, multi-agent, and hybrid categories — across 4 LLMs and 4 datasets finds no single method consistently dominates. The unified evaluation framework addresses fragmentation in prior black-box UE work where internal logits and hidden states are unavailable.