CIPHER Cross-Record Inference - MIMIC (Least-to-Most): leaderboard

Metric: Harmonic answer score (%; per question the harmonic mean of token overlap with the reference and Gemini 2.5 Flash semantic acceptance, zero when rejected, averaged; MIMIC clinical records with redacted narratives; the full support-controlled context of 39-42 records including distractors; least-to-most decomposition prompting; greedy decoding). Source: arxiv.org. Saturation forecast: Around October 2027. 5 models tracked.

Top models

#ModelScore
1Llama 3.3 70B Instruct38.3
2Gemini 2.0 Flash37.3
3Qwen 3 Next 80B A3B36.5
4GPT-OSS-20B34.9
5GPT-OSS-120B34.6

Interactive version: theaggregate.ai/benchmark?slug=cipher-cross-record-inference-mimic-least-to-most · How It Works · Data refreshed daily, snapshot 2026-09-26.