GraphVerse - Traveling Salesman - VGR-Score: leaderboard
GraphVerse (arXiv 2608.06769v1, Table 1) visual graph reasoning score for a traveling-salesman tour over airports on edited graph images: 0.7 weight on reasoning-process correctness, checked by a GPT-5.1 judge against verbalized graph evidence, and 0.3 on answer correctness. 0-100 on the 80-item testmini split; higher is better.
Metric: Visual graph reasoning score (0-100; 0.7 process + 0.3 answer, GPT-5.1 verifier). Source: arxiv.org. Saturation forecast: Around 2028. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 48.5 | #103 |
| 2 | Gemini 3 Pro | 42.1 | #82 |
| 3 | Kimi K2.5 (Thinking) | 41.7 | #152 (Kimi K2.5) |
| 4 | QVQ-72B-Preview | 37.3 | #458 |
| 5 | GLM-4.6V | 33.8 | #330 |
| 6 | GPT-5.2 | 33 | #115 |
| 7 | Qwen 3.5 4B | 31.5 | #514 |
| 8 | Qwen-VL-Max | 30 | #475 |
| 9 | InternVL3-38B | 28.4 | #413 |
| 10 | InternVL3-8B | 16.5 | #624 |
| 11 | Grok 4 Fast | 12.9 | #266 |
| 12 | O1 | 9.8 | #190 |
| 13 | Qwen 3.5 35B A3B | 8.5 | #268 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=graphverse-traveling-salesman-vgr-score · How It Works · Data refreshed daily, snapshot 2026-11-05.