SciTrue - Evidence Pairing: leaderboard
Honest per-sample development results on NTCIR SciClaimEval. Author runs include structured table data as an auxiliary text channel; official baseline runs and the PNG-only extension are kept on separate protocol boards. Claim validation is macro-F1, evidence pairing requires both paired claims correct, and attribution is Subtask 2 accuracy. This board is the authors’ structured-table-text protocol.
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Around April 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5 | 74.4 | #20 |
| 2 | GPT-5.5 | 69.9 | #33 |
| 3 | Gemma 4 31B (IT) | 69.9 | #179 |
| 4 | Claude Opus 4.8 | 67.6 | #40 |
| 5 | Qwen 3.6 35B A3B | 66.2 | #239 |
| 6 | Qwen 3.5 35B A3B | 61.1 | #268 |
| 7 | gemma-4-E4B-it | 52.6 | #577 |
| 8 | Qwen 3.5 9B | 48.3 | #391 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #6 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=scitrue-evidence-pairing · How It Works · Data refreshed daily, snapshot 2026-11-06.