Psi-Bench - Everyday Request Personalization: leaderboard
Personalization in the everyday-request scenario of Psi-Bench (2026): how well the model tailors its responses 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 response score (how well responses are tailored to the hidden client profile), everyday requests (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 February 2028. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.1 | 6.12 | #229 |
| 2 | Qwen 3 Next 80B A3B | 5.77 | #757 |
| 3 | DeepSeek V4 Pro | 5.15 | #188 |
| 4 | Gemini 3.1 Pro (Preview) | 5.14 | #94 |
| 5 | Gemini 3 Flash | 5.1 | #164 |
| 6 | Qwen 3 32B | 4.66 | #960 |
| 7 | GPT-5 Mini | 4.42 | #348 |
| 8 | Grok 4 Fast | 4.34 | #441 |
| 9 | DeepSeek V3.2 | 4.32 | #406 |
| 10 | Qwen 3 8B | 3.63 | #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-everyday-request-personalization · How It Works · Data refreshed daily, snapshot 2026-11-13.