MARINER - VQA: leaderboard

Maritime visual question answering on MARINER (2026), with multiple-choice questions built from verified image metadata and audited by maritime annotators. Accuracy in percent on eight task types (attributes, equipment grounding, text, perspective, spatial relations, environment and operational status), averaged over the types; higher is better.

Metric: Accuracy (%) on MARINER maritime visual question answering (mean of the eight task types in perception, space and reasoning; multiple-choice questions generated from verified image metadata and audited by maritime annotators); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 16 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 2.5 Pro73.97#280
2GPT-4.173.77#506
3InternVL3-38B69.29#948
4InternVL3-78B69.05#765
5GPT-4o68.91#754
6Qwen 2.5 VL 72B Instruct68.79#849
7LLaVA-OneVision-72B67.13#1782
8Qwen 2.5 VL 7B Instruct65.63#1919
9InternVL3-8B64.74#1795
10Gemini 2.5 Flash64.52#466
11Qwen 2.5 VL 32B Instruct63.85#1065
12MiniCPM-V-2.661.57#3311
13Qwen 2.5 VL 3B Instruct60.67#2720
14LLaVA-OneVision-7B60.47#3002
15InternVL2-8B60.08#3268

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=mariner-vqa · How It Works · Data refreshed daily, snapshot 2026-11-22.