Avalon-ToM-Bench - Average: leaderboard
Avalon-ToM-Bench (arXiv 2608.09638v1, Table 1): theory of mind in The Resistance: Avalon, where models judge perspective-constrained true/false statements about players' knowledge and motives from game state and public history. Accuracy on the 408-item balanced set (chance 50%), macro-averaged over four abilities; zero-shot. Higher is better.
Metric: Accuracy (%; macro-average of four abilities, balanced true/false). Source: arxiv.org. Saturation forecast: Around February 2027. 27 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 89.67 | #159 |
| 2 | GPT-5 | 88.5 | #105 |
| 3 | Gemini 2.5 Flash | 86.75 | #255 |
| 4 | Qwen 3 4B 2507 (Thinking) | 80.8 | #527 |
| 5 | GPT-5 Mini | 80.37 | #193 |
| 6 | Qwen 3 8B (Thinking) | 79.51 | #678 (Qwen 3 8B) |
| 7 | Qwen 3 14B (Reasoning) | 77.25 | #542 (Qwen 3 14B) |
| 8 | GPT-4o Mini | 75.04 | #598 |
| 9 | Qwen 3 4B (Reasoning) | 74 | #834 (Qwen 3 4B) |
| 10 | Qwen 3.5 9B (Thinking) | 73.82 | #391 (Qwen 3.5 9B) |
| 11 | Phi-4 | 73.61 | #707 |
| 12 | Gemma 3 27B (IT) | 71.51 | #528 |
| 13 | GPT-OSS-120B | 71.05 | #346 |
| 14 | Mistral 7B Instruct (v0.3) | 70.91 | #1255 |
| 15 | Llama 3 8B Instruct | 69.98 | #1116 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=avalon-tom-bench-average · How It Works · Data refreshed daily, snapshot 2026-11-05.