GroupToM-Bench - Desire: leaderboard
Metric: strict exact-match accuracy (%) over multi-select options (random 6.7) on Level 2 desire inference (latent psychological motives behind stated goals), multiple choice; 240 multimodal multi-agent scenarios (scene image plus dialogue); higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 11 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro | 77.1 |
| 2 | GPT-4o | 75.3 |
| 3 | GPT-5 | 74.1 |
| 4 | Claude Haiku 4.5 | 73.2 |
| 5 | GPT-5 Mini | 70.3 |
| 6 | GPT-5 Nano | 64.8 |
| 7 | InternVL3.5-8B | 60.7 |
| 8 | Qwen 2.5 VL 7B Instruct | 58.2 |
| 9 | Qwen 2 VL 7B | 54.9 |
Interactive version: theaggregate.ai/benchmark?slug=grouptom-bench-desire · How It Works · Data refreshed daily, snapshot 2026-09-29.