Persuasion (Lechmazur): leaderboard
Multi-turn persuasion benchmark where one LLM tries to shift another model's stated position on policy propositions. 15 models, 15 topics, 6,296 conversations. Measures average signed stance shift toward persuader's assigned side.
Metric: Average Persuasion Strength. Source: github.com. Saturation forecast: Around 2028. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 (High) | 1.71 | #79 (GPT-5.4) |
| 2 | Claude Opus 4.6 (Thinking, High) | 1.67 | #63 (Claude Opus 4.6) |
| 3 | Seed 2.0 Pro | 1.64 | #118 |
| 4 | Kimi K2.5 (Thinking) | 1.33 | #144 (Kimi K2.5) |
| 5 | Gemini 3.1 Pro (Preview) | 1.23 | #58 |
| 6 | GLM-5 | 1.22 | #141 |
| 7 | Grok 4.20 Beta (0309) (Reasoning) | 1.2 | #204 (Grok 4.20 Beta (0309)) |
| 8 | Qwen 3.5 397B A17B | 0.98 | #146 |
| 9 | MiniMax-M2.7 | 0.96 | #258 |
| 10 | Gemini 3.1 Flash Lite (Preview) | 0.81 | #194 |
| 11 | ERNIE 5.0 | 0.77 | #233 |
| 12 | DeepSeek V3.2 | 0.71 | #205 |
| 13 | MiMo-V2-Pro | 0.52 | #168 |
| 14 | Mistral Large 3 | 0.42 | #401 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=persuasion-lechmazur · How It Works · Data refreshed daily, snapshot 2026-10-20.