VB (Visibility Benchmark) - Confidence-Aware Accuracy: leaderboard
Metric: Confidence-aware accuracy with abstention (CAA, 0-1, shown times 100): a correct answer earns its stated confidence, a wrong one zero and ABSTAIN 0.25, averaged over the 300 headline items; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 76.9 | #333 |
| 2 | Gemini 3.1 Pro (Preview) | 76 | #54 |
| 3 | Gemini 2.5 Pro | 74.7 | #145 |
| 4 | GPT-5 | 62.2 | #91 |
| 5 | Claude Opus 4.5 | 58 | #79 |
| 6 | Gemma 3 12B | 54.3 | #666 |
| 7 | Claude 3.7 Sonnet | 50.8 | #241 |
| 8 | InternVL3-8B | 49.8 | #606 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=vb-visibility-benchmark-confidence-aware-accuracy · How It Works · Data refreshed daily, snapshot 2026-10-11.