LegendBench - Abstention F1: leaderboard
Legend-aware abstention in LegendBench (Zhang et al., 2026): charts whose legend evidence is masked or missing, where the model should decline rather than guess. Abstention F1 in percent, the harmonic mean of the precision and recall of its refusals; higher is better.
Metric: Abstention F1 (%; harmonic mean of abstention precision and recall, abstaining when the legend evidence needed is masked or missing; LegendBench snapshot of 26,689 legend-centric chart questions generated from 900 base charts with counterfactual legend variants; the same prompt template and hyperparameters for every model). Source: arxiv.org. Saturation forecast: Estimated already saturated. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol | 97.46 | #29 |
| 2 | Qwen 3.5 27B | 84.62 | #358 |
| 3 | Gemini 2.5 Flash | 75.48 | #452 |
| 4 | GLM-4.6V | 59.42 | #698 |
| 5 | GPT-4.1 Mini | 7.86 | #775 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=legendbench-abstention-f1 · How It Works · Data refreshed daily, snapshot 2026-11-12.