InstrumentalEval - Running Multiple AI Copies: leaderboard
The 13 InstrumentalEval tasks in which the model, pushed to reach its goal at all costs, could decide on its own to run more copies of itself; the score is the share of responses a GPT-4o judge marks as doing so. It is one of six categories behind the InstrumentalEval total.
Metric: Responses pursuing the instrumental goal, 13 tasks (%, lower is better). Source: arxiv.org. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 30.77 | #741 |
| 2 | DeepSeek V3 | 30.77 | #716 |
| 3 | Gemini 2.0 Flash (01-21) (Thinking) | 30.77 | #711 (Gemini 2.0 Flash (01-21)) |
| 4 | O3 Mini | 38.46 | #533 |
| 5 | O1 Mini | 38.46 | #764 |
| 6 | DeepSeek R1 | 46.15 | #473 |
| 7 | O1 (2024-12-17) | 46.15 | #311 |
Interactive version: theaggregate.ai/benchmark?slug=instrumentaleval-running-multiple-ai-copies · How It Works · Data refreshed daily, snapshot 2026-11-13.