GEB-Bench - Story-Theorem Matching: leaderboard
Metric: Accuracy (%; 4-way, chance 25%). Source: arxiv.org. 37 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 | 82.6 |
| 2 | Gemini 3.1 Pro (Preview) | 74.4 |
| 3 | Kimi K2.5 | 74.4 |
| 4 | GPT-5.2 | 73.6 |
| 5 | Claude Opus 4.8 | 73.6 |
| 6 | Gemma 4 31B | 68.6 |
| 7 | GLM-4.6V | 57.9 |
| 8 | DeepSeek V3.2 | 47.1 |
| 9 | Llama 4 Scout | 45.5 |
| 10 | Qwen 3 VL 4B | 41.3 |
| 11 | ERNIE 4.5 VL 28B A3B | 41.3 |
| 12 | InternVL3-8B | 38 |
| 13 | Kimi-VL-A3B | 33.1 |
| 14 | Qwen 2 VL 7B | 32.2 |
| 15 | Pixtral-12B | 28.1 |
Interactive version: theaggregate.ai/benchmark?slug=geb-bench-story-theorem-matching · How It Works · Data refreshed daily, snapshot 2026-09-19.