FDARxBench - Citation F1 (Multi-Hop): leaderboard
Metric: Macro F1 (x 100) between cited and gold provenance passage ids on multi-hop questions from FDARxBench's expert-guided QA items grounded in 700 FDA prescription drug labels, where the whole drug label is given as passage-indexed context and the model answers with cited passage ids; higher is better. Source: arxiv.org. Saturation forecast: Around 2034. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 45.8 | #60 |
| 2 | Llama 3.3 70B Instruct | 43.3 | #520 |
| 3 | Qwen 3 32B | 43.3 | #424 |
| 4 | GPT-5.1 | 40.6 | #131 |
| 5 | Claude Sonnet 4.5 | 38.3 | #138 |
| 6 | GPT-5.2 | 38.3 | #105 |
| 7 | GPT-4o Mini | 37.8 | #588 |
| 8 | Qwen 3 14B | 36.2 | #524 |
| 9 | Llama 3.1 8B Instruct | 33.3 | #1018 |
| 10 | Ministral-3-14B-Instruct-2512 | 32.8 | #590 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=fdarxbench-citation-f1-multi-hop · How It Works · Data refreshed daily, snapshot 2026-10-11.