ResearchQA (Open Paper) - Refusal Correctness: leaderboard
ResearchQA (2026) asks citation-grounded questions about single open-access scientific papers, run here through the Open Paper chat-with-paper harness on a 100-question sample. Refusal correctness is the percent of false-premise questions answered with a refusal or with only real, grounded citations; higher is better.
Metric: Refusal correctness (%; share of the adversarial false-premise questions in the sample answered with a refusal or with only grounded citations, citation-grounded answers from the Open Paper chat-with-paper harness to a 100-row evenly spaced sample of 6,211 single-paper questions over 494 open-access papers, citations checked by a normalizing substring matcher against the extracted paper text; higher is better). Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 87 | #94 |
| 2 | GPT-4.1 | 87 | #495 |
| 3 | GPT-5.4 | 78.3 | #123 |
| 4 | GLM-4.7 | 78.3 | #366 |
| 5 | Claude Haiku 4.5 | 73.9 | #568 |
| 6 | Claude Opus 4.7 | 73.9 | #79 |
| 7 | Gemini 3 Flash (Preview) | 65.2 | #144 |
| 8 | GPT-OSS-120B | 47.8 | #700 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=researchqa-open-paper-refusal-correctness · How It Works · Data refreshed daily, snapshot 2026-11-13.