RealityTest: leaderboard
AI systems are increasingly deployed in conversational settings where users may be uncertain whether they are speaking with a human or an AI. Despite mounting regulatory attention to this known safety risk, existing evaluations of AI disclosure are typically English-only, based on machine-generated questions, and restricted to text. We present RealityTest to comprehensively test whether AI systems
Metric: Text disclosure probability (self-reported). Source: benchmarklist.com. Saturation forecast: Around August 2027. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Haiku 4.5 | 92.3 | #280 |
| 2 | GPT-5.1 | 85.6 | #135 |
| 3 | Claude Opus 4.6 | 66 | #63 |
| 4 | Kimi K2 | 55.2 | #243 |
| 5 | Claude Sonnet 4 | 51.5 | #199 |
| 6 | Grok 4 | 41.3 | #181 |
| 7 | Gemma 4 31B | 31.3 | #200 |
| 8 | Gemini 3.1 Pro (Preview) | 30.5 | #58 |
| 9 | O4 Mini | 28.3 | #180 |
| 10 | DeepSeek R1 | 23.4 | #251 |
| 11 | Gemini 3 Flash | 22.7 | #99 |
| 12 | DeepSeek V3 | 14.7 | #321 |
| 13 | Gemini 2.5 Pro | 12.8 | #152 |
| 14 | GPT-4o | 12.6 | #343 |
| 15 | Llama 3.3 70B | 12.5 | #579 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=realitytest · How It Works · Data refreshed daily, snapshot 2026-10-20.