EPPCMinerBen (2-Shot) - Evidence Extraction: leaderboard
Metric: Micro-averaged F1 (%) of EPPC evidence extraction (the minimal supporting span for each code-subcode pair, matched to the expert span by containment or token Jaccard similarity of at least 0.6), two in-context examples, mean over seeded runs, on 1,933 expert-annotated sentences from 752 Yale New Haven Health patient-portal secure messages; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.1 70B Instruct | 82.84 | #548 |
| 2 | Llama 3.3 70B Instruct | 78.84 | #520 |
| 3 | DeepSeek R1 Distill Qwen 32B | 77.24 | #640 |
| 4 | Qwen2.5-14B-Instruct-1M | 74 | #566 |
| 5 | DeepSeek R1 Distill Llama 70B | 73.9 | #582 |
| 6 | Gemma 2 27B (IT) | 67.31 | #633 |
| 7 | Qwen 2.5 1.5B Instruct | 50.1 | #1388 |
| 8 | Llama 3.1 8B Instruct | 36.42 | #1018 |
| 9 | Llama 3.2 3B Instruct | 32.47 | #1321 |
| 10 | Llama 3.2 1B Instruct | 6.84 | #1518 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=eppcminerben-2-shot-evidence-extraction · How It Works · Data refreshed daily, snapshot 2026-10-11.