SALAD-Bench
Explore score history ↗21,318 prompts (arXiv:2608.17183, §3.1). MD-Judge-v0.1 natively produces an unsafe probability plus a binary safe/unsafe judgment, not a ternary score: the paper derives ternary 0 / 0.5 / 1 outcomes with model-specific Otsu thresholds (§3.2) to compare SALAD with the HELM-style suites, so the ternary scale is the paper's derivation, not SALAD-Bench's native protocol, and a future native MD-Judge score must not be read as ternary. No per-model scores filed: the paper reports ambiguity and rank instability across suites rather than benchmark scores.
- Evaluates
- security
- Openness
- unknown
- Importer
- Model reports
No reported scores are on record for this benchmark yet.