Avalon-ToM-Bench - Intention Attribution: leaderboard
Metric: Accuracy (%). Source: arxiv.org. 27 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.5 Pro | 100 |
| 2 | GPT-5 | 96.67 |
| 3 | Gemini 2.5 Flash | 95 |
| 4 | GPT-5 Mini | 86.67 |
| 5 | Qwen 3 4B (Reasoning) | 86.27 |
| 6 | Qwen 3 8B (Thinking) | 85.11 |
| 7 | Qwen 3 4B 2507 (Thinking) | 85 |
| 8 | GPT-4o Mini | 83.33 |
| 9 | Qwen 3 14B (Reasoning) | 83.05 |
| 10 | GPT-OSS-120B | 81.67 |
| 11 | Llama 3 8B Instruct | 81.67 |
| 12 | Qwen 3.5 9B (Thinking) | 80 |
| 13 | Llama 3.2 3B Instruct | 78.43 |
| 14 | Phi-4 | 78.33 |
| 15 | Phi-3.5-mini-instruct | 78.33 |
Interactive version: theaggregate.ai/benchmark?slug=avalon-tom-bench-intention-attribution · How It Works · Data refreshed daily, snapshot 2026-09-19.