ConflictBench (Text-Only): leaderboard
Metric: Task success rate (%): share of scenarios in which the agent's actions reach a human-favorable terminal outcome in the environment, over ConflictBench's 150 multi-turn human-AI conflict scenarios (51 self-preservation versus human safety, 47 resource prioritization and 52 deceptive alignment episodes, adapted from PacifAIst), macro-averaged over the three categories, temperature 0; text-only setting: the agent sees only the text simulation engine, at most 14 turns; higher is better. Source: arxiv.org. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 65.51 | #91 |
| 2 | Qwen Plus | 59.71 | #370 |
| 3 | GPT-4o | 59.57 | #333 |
| 4 | DeepSeek V3 | 53.64 | #312 |
| 5 | Qwen 3 VL 30B A3B Instruct | 53.57 | #365 |
| 6 | Gemini 2.5 Flash | 44.35 | #237 |
| 7 | GPT-4o Mini | 35.1 | #588 |
Interactive version: theaggregate.ai/benchmark?slug=conflictbench-text-only · How It Works · Data refreshed daily, snapshot 2026-10-11.