MultivationBench: leaderboard
MultivationBench (HKUST, Findings of EMNLP 2026) tests motivation reasoning about characters in story-driven visual narratives under the Maslow hierarchy of needs and the 16 Reiss basic desires, from the accumulated story images and text. The score is exact-match accuracy of the predicted motivation set in percent, on the common subset of tasks every model answered; higher is better.
Metric: Exact-match accuracy (%, predicted option set equals the gold set) of multi-label motivation predictions for visually grounded character behaviors in sequential visual narratives, reasoning over the accumulated story images and text (Maslow 8-level needs and Reiss 16 basic desires, definition and practical-motivation tasks), zero-shot at temperature 0; multimodal input, common subset of 14,180 tasks from 1,000 stories that every model answered; higher is better. Source: arxiv.org. Saturation forecast: Around 2032. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Phi-4 Multimodal Instruct | 39.22 | #3882 |
| 2 | Gemini 3 Flash | 36.09 | #164 |
| 3 | Llama 4 Scout | 33.48 | #1776 |
| 4 | Grok 4.1 Fast | 32.53 | #430 |
| 5 | Llama 4 Maverick | 30.28 | #1099 |
| 6 | O4 Mini | 27.29 | #353 |
| 7 | Nemotron Nano 12B v2 VL | 8.58 | #2231 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=multivationbench · How It Works · Data refreshed daily, snapshot 2026-11-13.