GEB-Bench - Story-Theorem Matching: leaderboard
GEB-Bench (arXiv 2608.04111v2, Tables 2-3): four-way matching of a story to the theorem that states the same motif, with decoys from other motifs (text input, all models). Judge-free exact-match multiple choice with greedy decoding; accuracy in percent, chance 25%; higher is better.
Metric: Accuracy (%; 4-way, chance 25%). Source: arxiv.org. Saturation forecast: Around February 2027. 37 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 82.6 | #33 |
| 2 | Gemini 3.1 Pro (Preview) | 74.4 | #64 |
| 3 | Kimi K2.5 | 74.4 | #152 |
| 4 | GPT-5.2 | 73.6 | #115 |
| 5 | Claude Opus 4.8 | 73.6 | #39 |
| 6 | Gemma 4 31B | 68.6 | #212 |
| 7 | GLM-4.6V | 57.9 | #330 |
| 8 | DeepSeek V3.2 | 47.1 | #210 |
| 9 | Llama 4 Scout | 45.5 | #648 |
| 10 | ERNIE 4.5 VL 28B A3B | 41.3 | |
| 11 | InternVL3-8B | 38 | #624 |
| 12 | Qwen 2 VL 7B | 32.2 | #743 |
| 13 | Pixtral-12B | 28.1 | #805 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=geb-bench-story-theorem-matching · How It Works · Data refreshed daily, snapshot 2026-11-05.