MARINER - VQA Environmental State Inference: leaderboard
Metric: Accuracy (%) on MARINER maritime visual question answering (environmental state questions; multiple-choice questions generated from verified image metadata and audited by maritime annotators); higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 16 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.5 Pro | 92.38 |
| 2 | InternVL3-78B | 92.31 |
| 3 | InternVL3-38B | 91.82 |
| 4 | Qwen 2.5 VL 7B Instruct | 91.33 |
| 5 | Qwen 2.5 VL 72B Instruct | 91.26 |
| 6 | InternVL3-8B | 90.85 |
| 7 | MiniCPM-V-2.6 | 90.78 |
| 8 | Gemini 2.5 Flash | 90.08 |
| 9 | GPT-4.1 | 90.01 |
| 10 | InternVL2-8B | 88.61 |
| 11 | GPT-4o | 85.32 |
| 12 | Qwen 2.5 VL 32B Instruct | 74.63 |
Interactive version: theaggregate.ai/benchmark?slug=mariner-vqa-environmental-state-inference · How It Works · Data refreshed daily, snapshot 2026-10-07.