OmniMapBench - Multi-Step Reasoning: leaderboard
Metric: Level 3 (multi-step reasoning, 853 questions): strict exact-match accuracy on manually annotated single-choice, multiple-choice and ordering questions over map documents, chain-of-thought prompting (%); higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 25 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 62.88 |
| 2 | Qwen 3.5 397B A17B | 60.37 |
| 3 | Qwen 3.5 Plus | 57.47 |
| 4 | Kimi K2.5 | 56.71 |
| 5 | Qwen 3.5 27B | 53.97 |
| 6 | Qwen 3.5 122B A10B | 53.54 |
| 7 | Qwen 3.5 35B A3B | 49.93 |
| 8 | Qwen 3.5 Flash | 48.62 |
| 9 | GPT-5 | 48.53 |
| 10 | Gemini 2.5 Pro | 46.31 |
| 11 | GPT-5 Mini | 45.13 |
| 12 | Gemma 4 31B (IT) | 40.68 |
| 13 | GPT-4.1 | 37.87 |
| 14 | GLM-4.5V | 35.8 |
| 15 | GPT-4.1 Mini | 35.64 |
Interactive version: theaggregate.ai/benchmark?slug=omnimapbench-multi-step-reasoning · How It Works · Data refreshed daily, snapshot 2026-09-29.