Psi-Bench - Everyday Request: leaderboard
Everyday requests in Psi-Bench (2026): how well the model changes the behaviour of a simulated, persona-profiled client on profile-grounded everyday requests, rated 1 to 9 by a DeepSeek-V3.2 judge that also plays the client. Higher is better.
Metric: Effect on the client's behaviour, profile-grounded 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. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.1 | 5.88 | #229 |
| 2 | Qwen 3 Next 80B A3B | 5.85 | #757 |
| 3 | Qwen 3 32B | 4.21 | #960 |
| 4 | Qwen 3 8B | 4.12 | #1830 |
| 5 | GPT-5 Mini | 4.11 | #348 |
| 6 | Gemini 3 Flash | 3.98 | #164 |
| 7 | DeepSeek V4 Pro | 3.92 | #188 |
| 8 | Gemini 3.1 Pro (Preview) | 3.66 | #94 |
| 9 | DeepSeek V3.2 | 3.34 | #406 |
| 10 | Grok 4 Fast | 3.05 | #441 |
Interactive version: theaggregate.ai/benchmark?slug=psi-bench-everyday-request · How It Works · Data refreshed daily, snapshot 2026-11-13.