Multicultural Spyfall: leaderboard
Multicultural Spyfall (2026) has models play five-player Spyfall in English, Indonesian, Chinese and Egyptian Arabic, with generic locations or culture-specific locations and foods as the secret, one model playing the spy and another the four non-spies. The score is a model's win rate (%) over the 9,000 released games against the other evaluated models, in both roles; higher is better.
Metric: Win Rate (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 70.13 | #262 |
| 2 | Gemini 2.5 Flash | 66 | #452 |
| 3 | Qwen 3 30B A3B 2507 (Thinking) | 51.93 | #759 |
| 4 | Gemma 3 12B (IT) | 49.93 | #1849 |
| 5 | Qwen 3 8B | 44.3 | #1838 |
| 6 | Llama 3.1 8B Instruct | 17.7 | #4277 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=multicultural-spyfall · How It Works · Data refreshed daily, snapshot 2026-11-12.