Four-Factor Integrity Breakdown & Primary-Source Evidence
Access Model: Strictly Private
"Artificial Analysis evaluates a private held-out set of the two hardest tiers (expert-tier clinical synthesis and compound, multi-part reasoning): 60 questions, each run with 3 repeats. This private set is separate from the publicly released dataset"
Originality
Weight 25%
10/10
Reuse Profile: All Original (0% reuse)
"Each case is a realistic synthetic medical record of roughly 25,000 to 64,000 tokens. Unless a question sets its own format, we ask models to answer each part in one paragraph and cite the supporting evidence from the record."
Freshness
Weight 20%
10/10
Initial Release: 2026-06Latest Refresh: 2026-08Rolling Updates: No
"[2026/08/31] We are excited to release MLCR v1.1 that comes with some improvements to judging calibration and a release with Artificial Analysis"
Integrity: Independent Offline (Leak-Verified)Sandbox: not applicable single turn
"Responses are scored by a panel of three LLM judges (Claude Opus 5.5, GPT-6.1 Sol, and Grok 4.7, each at medium reasoning effort) against expert-annotated ground truth and the source medical record."
Frontier Model Standings on MLCR-AA
Top 15 active models tracked
Benchmark maintainers can embed this live SVG badge in GitHub READMEs or documentation:
[](https://benchaudit.org/benchmarks/mlcr-aa/)