OmniMapBench - Single-Step Reasoning: leaderboard
Metric: Level 2 (single-step reasoning, 802 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) | 81.53 |
| 2 | Qwen 3.5 397B A17B | 79.01 |
| 3 | Qwen 3.5 Plus | 77.64 |
| 4 | Qwen 3.5 27B | 77.1 |
| 5 | Qwen 3.5 122B A10B | 76.52 |
| 6 | Kimi K2.5 | 74.8 |
| 7 | Qwen 3.5 Flash | 73.87 |
| 8 | Qwen 3.5 35B A3B | 72.27 |
| 9 | Gemini 2.5 Pro | 70.7 |
| 10 | GPT-5 Mini | 62.59 |
| 11 | GPT-5 | 61.47 |
| 12 | Gemma 4 31B (IT) | 58.35 |
| 13 | GLM-4.5V | 54.24 |
| 14 | Claude Sonnet 4.5 | 52.63 |
| 15 | Gemma 4 26B A4B (IT) | 49.75 |
Interactive version: theaggregate.ai/benchmark?slug=omnimapbench-single-step-reasoning · How It Works · Data refreshed daily, snapshot 2026-09-29.