FraudSMSWalker: Benchmarking Agentic Large Language Models for SMS-to-Webpage Fraud Detection: leaderboard
FraudSMSWalker evaluates agentic language models on URL-masked SMS-to-webpage fraud judgments using text and sanitized webpage evidence. It includes bilingual chains where raw URLs, hosts, domains, IPs, redirects, and reputation metadata are withheld. The pinned metric is accuracy, measuring how often the model’s fraud judgment matches the benchmark labels under masked browser-agent protocols. Source: BenchmarkList’s attributed result table.
Metric: Accuracy (%). Source: benchmarklist.com. Saturation forecast: Not forecast. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.6 Plus | 50.93 | #124 |
| 2 | Seed 2.0 Pro | 49.5 | #122 |
| 3 | Kimi K2.6 | 49.36 | #113 |
| 4 | GLM-5-Turbo | 47.93 | #121 |
| 5 | GPT-5.5 | 46.78 | #33 |
| 6 | DeepSeek V4 Pro | 46.78 | #112 |
| 7 | Gemini 3.1 Flash Lite | 46.35 | #200 |
| 8 | Claude Sonnet 4.6 (Thinking) | 42.06 | #94 (Claude Sonnet 4.6) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=fraudsmswalker-benchmarking-agentic-large-language-models-for-sms-to-webpage-fraud-detection · How It Works · Data refreshed daily, snapshot 2026-11-05.