OI-Bench - Threat Framing - Injected Accuracy: leaderboard
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Around September 2027. 12 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Haiku 4.5 | 85.78 |
| 2 | Grok 4.1 | 81.26 |
| 3 | GPT-5 | 74.29 |
| 4 | Llama 4 Scout | 73.82 |
| 5 | Qwen 3 235B A22B | 73.72 |
| 6 | GPT-5 Mini | 71.93 |
| 7 | DeepSeek R1 | 71.82 |
| 8 | Llama 4 Maverick | 68.1 |
| 9 | DeepSeek V3.2 | 67.55 |
| 10 | Qwen 3 8B | 63.55 |
| 11 | Gemini 2.5 Pro | 55.46 |
| 12 | Gemini 2.5 Flash Lite | 48.36 |
Interactive version: theaggregate.ai/benchmark?slug=oi-bench-threat-framing-injected-accuracy · How It Works · Data refreshed daily, snapshot 2026-09-25.