RiskChainBench - Web Investigation: leaderboard
Evidence-grounded web investigation of platform-abuse websites, scored as Task 2 accuracy over 600 websites given correct routing. Higher is better. Source: arxiv.org/abs/2609.16900.
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Not forecast. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol | 62.8 | #22 |
| 2 | GPT-5.2 | 57.3 | #115 |
| 3 | Kimi K2.5 | 55.2 | #152 |
| 4 | GPT-5.4 | 53.7 | #83 |
| 5 | Qwen 3.6 Plus | 51.7 | #124 |
| 6 | Claude Opus 4.8 | 49.8 | #39 |
| 7 | Kimi K3 | 47.5 | #35 |
| 8 | Kimi K2.6 | 41.5 | #113 |
| 9 | Claude Sonnet 5 | 26.3 | #65 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=riskchainbench-web-investigation · How It Works · Data refreshed daily, snapshot 2026-11-05.