HelpBench - Scams: leaderboard
HelpBench (2026) asks LLMs real-world digital privacy, safety and security help-seeking questions and scores each answer against an expert-written rubric of factual and delivery criteria, applied by a Gemini 2.5 Pro auto-rater and averaged over five responses. This board is the rubric score in percent on questions about scams, fraud, spam and suspicious artifacts; higher is better.
Metric: Scams topic (scams, fraud, spam and suspicious artifacts) rubric score (%): points for met positive criteria and avoided negative criteria (factual criteria written per question by privacy, safety and security experts, plus shared delivery criteria) over the maximum points, applied by a Gemini 2.5 Pro auto-rater at temperature 0; chat versions with default parameters, mean of five responses per question; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Chat | 93 | #444 |
| 2 | Grok 4 | 91 | #307 |
| 3 | GPT-4.1 | 90 | #495 |
| 4 | Claude Opus 4.6 | 90 | #102 |
| 5 | Claude Sonnet 4 | 90 | #400 |
| 6 | Qwen 3.6 Plus | 90 | #222 |
| 7 | GPT-5.3 | 90 | #197 |
| 8 | Grok 4.20 | 89 | #297 |
| 9 | Claude Sonnet 4.6 | 88 | #161 |
| 10 | DeepSeek V3.2 (Non-reasoning) | 88 | #406 (DeepSeek V3.2) |
| 11 | Gemini 2.5 Pro | 87 | #266 |
| 12 | Gemini 3 Flash | 87 | #164 |
| 13 | GLM-5.1 | 86 | #179 |
| 14 | DeepSeek V3.1 (Non-reasoning) | 86 | #497 (DeepSeek V3.1) |
| 15 | Gemini 3.1 Pro (Preview) | 85 | #94 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=helpbench-scams · How It Works · Data refreshed daily, snapshot 2026-11-13.