Benchmark Almanac

MarineEval

Gemini-2.0-Flash-Vision 55.1% tops its leaderboard Grok 2 Vision (1212) 50.4% leads the models we track
Not in model cards

Models answer 2,000 questions about marine images and documents. The authors combine answer checks with model judgments. They report total accuracy and mean accuracy across seven tasks.

Passes tests 2 and 4. The authors publish scores for 17 vision-language models. The USI-Brain team scored 12 categories with different judges.

Released
November 2025
Kind
General visual questions
Level
University
Scored
Graded by a model
Answers
Open-ended
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Images, Documents → Text
Interaction
Single turn
Items from
Written by experts
Refresh
One fixed set
Run by
Anyone
Strongest model tested
Claude 3.7 Sonnet #119 on the capability index
Last model added
April 2025
Scores refreshed
November 20th, 2026
17
Entries on its own leaderboard, 8 models we track

Gemini-2.0-Flash-Vision tops the leaderboard with 55.1% Grok 2 Vision (1212) leads with 50.4% among models released by April 2025

Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is Grok 2 Vision (1212), at 50.4%. This view shows only models we track, each at its best.

The newest model with a score here came out in April 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 Gemini-2.0-Flash-Vision 55.1%
  2. 2 Grok 2 Vision (1212) 50.4%
  3. 3 Claude 3.7 Sonnet 48.9%
  4. 4 InternVL3-38B 47.5%
  5. 5 GPT-4o-Vision 45.6%
  6. 6 LLaVA-Next-Qwen 44.8%
  7. 7 LLaVA-1.6 Hermes-Yi 44.7%
  8. 8 LLaVA-1.6 Vicuna 44.7%
  9. 9 InternVL2.5-4B 42.5%
  10. 10 Qwen-VL-Plus 42.4%
  11. 11 InternVL2-8B 41.4%
  12. 12 InternLM-XComposer2.5-7B 41.1%
  13. 13 InternVL2.5-26B 38.6%
  14. 14 LLaVA-Next 37.5%
  15. 15 Mini-Monkey 34.5%
  16. 16 DeepSeek-VL-chat 25%
  17. 17 OpenFlamingo 17.6%

Source: Official MarineEval leaderboard

#EntryScoreSettingDateSource
1 Gemini-2.0-Flash-Vision closed weights 55.1% Official leaderboard
2 Grok 2 Vision (1212) 50.4% Dec 2024 Official leaderboard
3 Claude 3.7 Sonnet 48.9% Feb 2025 Official leaderboard
4 InternVL3-38B 47.5% Apr 2025 Official leaderboard
5 GPT-4o-Vision closed weights 45.6% Official leaderboard
6 LLaVA-Next-Qwen open weights, 32B 44.8% Official leaderboard
7 LLaVA-1.6 Hermes-Yi open weights, 34B 44.7% Official leaderboard
8 LLaVA-1.6 Vicuna open weights, 7B 44.7% Official leaderboard
9 InternVL2.5-4B 42.5% Dec 2024 Official leaderboard
10 Qwen-VL-Plus 42.4% Jan 2024 Official leaderboard
11 InternVL2-8B 41.4% Jul 2024 Official leaderboard
12 InternLM-XComposer2.5-7B 41.1% Jul 2024 Official leaderboard
13 InternVL2.5-26B 38.6% Dec 2024 Official leaderboard
14 LLaVA-Next open weights, 8B 37.5% Official leaderboard
15 Mini-Monkey open weights, 2B 34.5% Official leaderboard
16 DeepSeek-VL-chat open weights, 1.3B 25% Official leaderboard
17 OpenFlamingo open weights, 2B 17.6% Official leaderboard
#ModelScoreSettingDateSource
1 Grok 2 Vision (1212) 50.4% Dec 2024 Official leaderboard
2 Claude 3.7 Sonnet 48.9% Feb 2025 Official leaderboard
3 InternVL3-38B 47.5% Apr 2025 Official leaderboard
4 InternVL2.5-4B 42.5% Dec 2024 Official leaderboard
5 Qwen-VL-Plus 42.4% Jan 2024 Official leaderboard
6 InternVL2-8B 41.4% Jul 2024 Official leaderboard
7 InternLM-XComposer2.5-7B 41.1% Jul 2024 Official leaderboard
8 InternVL2.5-26B 38.6% Dec 2024 Official leaderboard

Rankings by part

We read 7 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Behavior and Trait Extraction, Conservation and Threat Analysis and Document Interpretation. Grok 2 Vision (1212) leads 4 of them, and 3 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

Behavior and Trait Extraction

8 models we track have a score here.

  1. 1 Grok 2 Vision (1212) 77.6%
  2. 2 InternVL3-38B ↑1 74.1%
  3. 3 Claude 3.7 Sonnet ↓1 68.2%
  4. 4 InternVL2.5-4B 65.2%
  5. 5 InternLM-XComposer2.5-7B ↑2 64.2%
  6. 6 InternVL2-8B 55.2%
  7. 7 Qwen-VL-Plus ↓2 52.2%
  8. 8 InternVL2.5-26B 35.3%

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

xAI
Own leaderboard Not tested since Nov 2025

Source: Official MarineEval leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2024 2025 Near the ceiling (85%) Grok 2 Vision (1212) InternVL2.5-4B Qwen-VL-Plus 0% 20% 40% 60% 80% 100% 2024 2025 Near the ceiling (85%) Grok 2 Vision (1212) InternVL2.5-4B

Source: Official MarineEval leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Vision

All 636 in Vision
  • MMMU In use 11.5K college-level multimodal questions across 30 subjects. Multiple-choice and open accuracy. Qwen 3.6 Plus 86%
  • MMMU-Pro In use Harder MMMU: college-level multimodal questions with more options and vision-only inputs to remove text shortcuts. GPT-5.6 Sol 84.6%
  • CharXiv In use Chart understanding on real arXiv figures. Reasoning questions require synthesizing information from complex charts. Claude Fable 5 93.5%
  • MathVista In use Math reasoning in visual contexts (charts, geometry, puzzles) from 31 datasets. Answer accuracy. Seed2.1 Pro 90.7%
  • MATH-Vision (MathVision) In use 3,040 competition math problems with visual context. Answer accuracy. Qwen 3.8 Max 97.7%
  • MMStar In use Models answer 1,500 human-curated image questions that cannot be solved without the image, across 6 core capabilities. Scored by accuracy. Qwen 3.8 Max 85.9%

Data audit

9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 19 months ago, in April 2025, for InternVL3-38B. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (17 entries).
  • Pass Models scored 8 models we track, plus 9 other entries on its leaderboard.
  • Warning Newest score The newest came 19 months ago, in April 2025, for InternVL3-38B.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Claude 3.7 Sonnet (#119).
  • Pass Headroom The best is 50.4%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on marineeval.hkustvgd.com: 8 rows for models we track and 9 other entries at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check