{"record": {"aliases": ["AirBench", "AirBench 2024"], "artifacts": [], "audits": [], "categories": ["security"], "description": {"en": "5,694 prompts across 314 risk categories (arXiv:2608.17183, §3.1). Judged by GPT-4o under a ternary 0 / 0.5 / 1 rubric. The paper's own finding is that mean-score leaderboards over these suites are unstable under rubric ambiguity, so no per-model scores from the paper are filed here; this entry records the suite so future runs can attach scores."}, "documents": [], "evidence_summary": {"document_count": null, "model_count": null, "model_count_basis": null}, "key": "model-reports:airbench_2024", "modality": null, "name": "AirBench 2024", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"registry": "data/model_cards.yml", "score_archive": "data/benchmark_scores.yml", "source_url": null}, "publisher": null, "released": null, "released_reference": null, "schema_version": 1, "sizes": [], "slug": "airbench_2024", "source": "model_reports", "source_benchmark_id": "airbench_2024"}, "schema_version": 1, "scores_by_source": {}, "siblings": []}