← All benchmarks

Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)

Professional Work

Artificial Analysis' revised MLCR-AA methodology, first published between October 19 and 21, 2026: 60 private held-out expert-tier and compound questions from Wisedocs Medical Long Context Reasoning, over synthetic records of about 25,000 to 64,000 tokens, each run three times. A response passes when a majority of three judges (Claude Opus 5.5, GPT-6.1 Sol and Grok 4.7 at medium effort) each find it complete, naming every documentation gap on the way to the ground-truth conclusion, and free of claims that contradict the record. Empty responses fail. The earlier conciseness gate is gone and AA re-scored every model, so these results are not comparable with mlcr-aa.

Top models (higher is better)