Psi-Bench - Viewpoint Debate Personalization: leaderboard
Personalization in the viewpoint-debate scenario of Psi-Bench (2026): how well the model tailors its arguments to the hidden persona profile of a simulated client, rated 1 to 9 by a DeepSeek-V3.2 judge that also plays the client. Higher is better.
Metric: Personalized argumentation score (how well arguments are tailored to the hidden client profile), Change My View debates (1-9; DeepSeek-v3.2 judge, which also plays the persona-profiled client; three rounds after the client's opening message, client profile hidden from the model). Source: arxiv.org. Saturation forecast: Around April 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4 Pro | 5.62 | #188 |
| 2 | GPT-5.1 | 5.57 | #229 |
| 3 | Gemini 3.1 Pro (Preview) | 5.54 | #94 |
| 4 | Grok 4 Fast | 5.2 | #441 |
| 5 | GPT-5 Mini | 4.91 | #348 |
| 6 | DeepSeek V3.2 | 4.8 | #406 |
| 7 | Gemini 3 Flash | 4.73 | #164 |
| 8 | Qwen 3 32B | 4.34 | #960 |
| 9 | Qwen 3 Next 80B A3B | 4.1 | #757 |
| 10 | Qwen 3 8B | 3.56 | #1830 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=psi-bench-viewpoint-debate-personalization · How It Works · Data refreshed daily, snapshot 2026-11-13.