Psi-Bench - Everyday Request: leaderboard

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. Saturation forecast: Around 2032. 10 models tracked.

Top models

#ModelScore
1GPT-5.15.88
2Qwen 3 Next 80B A3B5.85
3Qwen 3 32B4.21
4Qwen 3 8B4.12
5GPT-5 Mini4.11
6Gemini 3 Flash3.98
7DeepSeek V4 Pro3.92
8Gemini 3.1 Pro (Preview)3.66
9DeepSeek V3.23.34
10Grok 4 Fast3.05

Interactive version: theaggregate.ai/benchmark?slug=psi-bench-everyday-request · How It Works · Data refreshed daily, snapshot 2026-09-29.