PICQ-drama (Multi-Task Prompting): leaderboard
The PICQ-drama missing-persona task (Su et al., University of Tokyo, 2026) under the paper's multi-task prompt, which has the model first summarize the choice situation and then generalize the personas it lists. Fidelity F1 in percent against the annotators' merged personas over 300 choice questions, matched by persona category and an NLI test; higher is better.
Metric: Fidelity F1 (%; up to five identified personas per question matched to the annotators' missing personas by category and NLI, multi-task prompt). Source: arxiv.org. Saturation forecast: Estimated already saturated. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 32B | 48 | #960 |
| 2 | Llama 3.1 8B Instruct | 41.8 | #4277 |
| 3 | Qwen 3 8B | 39.1 | #1838 |
| 4 | Llama 3.1 70B Instruct | 34.6 | #1541 |
| 5 | GPT-4.1 | 30 | #492 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=picq-drama-multi-task-prompting · How It Works · Data refreshed daily, snapshot 2026-11-12.