MARINER - Vessel Detection: leaderboard

Vessel detection by multimodal language models on MARINER (2026), a maritime vision benchmark of shore, shipboard, drone and open-source images. COCO-style average precision in percent over all vessel categories; higher is better.

Metric: Average precision of MARINER vessel detection by multimodal LLMs (COCO-style average precision over IoU 0.5 to 0.95, percent; all vessel categories); higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 13 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 2.5 Pro59.42#276
2InternVL3-38B27.06#939
3Qwen 2.5 VL 72B Instruct22.49#841
4Qwen 2.5 VL 32B Instruct17.13#1062
5Gemini 2.5 Flash16.94#463
6Qwen 2.5 VL 7B Instruct16.42#1902
7InternVL3-78B13.25#759
8MiniCPM-V-2.69.18#3267
9InternVL3-8B6.82#1782
10LLaVA-OneVision-72B6.76#1771
11LLaVA-OneVision-7B6.11#2981
12Qwen 2.5 VL 3B Instruct4.88#2713
13InternVL2-8B1.07#3227

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=mariner-vessel-detection · How It Works · Data refreshed daily, snapshot 2026-11-21.