SIGNPOST-Bench - Trap-Fit Rate: leaderboard
Metric: Adversarial predictions within 50 km of the injected location (%; lower is better). Source: arxiv.org. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.5 Pro | 6.5 |
| 2 | Gemini 3 Flash | 6.7 |
| 3 | Gemini 3.1 Pro (Preview) | 8 |
| 4 | Qwen 3 VL 30B A3B Instruct | 9.6 |
| 5 | Gemini 2.5 Flash | 10 |
| 6 | Claude Opus 4.6 | 11.4 |
| 7 | GPT-4o | 11.5 |
| 8 | GPT-5 | 11.6 |
| 9 | Claude Sonnet 4.6 | 11.8 |
| 10 | Kimi K2.5 | 13.1 |
| 11 | GPT-5.4 | 13.8 |
| 12 | Doubao-Seed-2.0-Pro-260215 | 15.9 |
| 13 | GPT-4o Mini | 16.7 |
| 14 | Qwen 3 VL 235B A22B Instruct | 16.9 |
| 15 | Grok 4 | 19.9 |
Interactive version: theaggregate.ai/benchmark?slug=signpost-bench-trap-fit-rate · How It Works · Data refreshed daily, snapshot 2026-09-19.