Benchmark Almanac

MarineEVT

Ours (EVT-R1) 48.9% tops its leaderboard Qwen3-VL-8B-Instruct 43.7% leads the models we track
Not in model cards

Models answer 2,000 held-out questions about underwater events, including species interactions, movement, and timing. The authors report accuracy across five reasoning categories.

The authors publish scores for eight general video models without tools. They also compare three of those models with tools on the same questions.

Released
June 2026
Kind
Video understanding
Level
University
Scored
Checked automatically
Answers
Short answer
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Video → Text, Actions
Interaction
Agentic
Environment
APIs and tools
Items from
Real work and use
Refresh
One fixed set
Run by
Anyone
Strongest model tested
Gemini 3 Flash #60 on the capability index
Last model added
December 2025
Scores refreshed
November 20th, 2026
18
Entries on its own leaderboard, 7 models we track

Ours (EVT-R1) tops the leaderboard with 48.9% Qwen3-VL-8B-Instruct leads with 43.7% among models released by December 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. Every run is in the next chart. 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 Qwen3-VL-8B-Instruct, at 43.7%. This view shows only models we track, each at its best.

The newest model with a score here came out in December 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 Ours (EVT-R1) Trained, Tools 48.9%
  2. 2 Qwen3-VL-8B (SFT) Trained, Tools 45.3%
  3. 3 Qwen3-VL-8B-Instruct No training, No tools 43.7%
  4. 4 InternVL3-8B No training, No tools 40.6%
  5. 5 GPT-5 mini No training, Tools 40.4%
  6. 6 VideoITG No training, No tools 39.6%
  7. 7 Gemini 3 Flash No training, No tools 37.6%
  8. 8 CoF No training, No tools 35.9%
  9. 9 Qwen3-VL-8B (GRPO†) Trained, Tools 35.6%
  10. 10 VideoLLAMA3-7B No training, No tools 31.6%
  11. 11 Grok 4.1 Fast reasoning, No training, No tools 30.6%
  12. 12 MaxInfo No training, No tools 30.4%
  13. 13 InternVL2-8B (PVC) No training, No tools 30.2%
  14. 14 AKS No training, No tools 28.7%
  15. 15 LLaVA-NeXT-Video-7B No training, No tools 25.6%
  16. 16 LLaVA-1.5-7B (VisionZip) No training, No tools 23.8%
  17. 17 Video-LLaVA-7B No training, No tools 21.9%
  18. 18 Qwen2.5-VL-7B (VisionZip) No training, No tools 20.4%

Source: Official MarineEVT leaderboard

#EntryScoreSettingDateSource
1 Ours (EVT-R1) open weights 48.9% Trained, Tools Official leaderboard
2 Qwen3-VL-8B (SFT) open weights 45.3% Trained, Tools Official leaderboard
3 Qwen3-VL-8B-Instruct 43.7% No training, No tools Oct 2025 Official leaderboard
4 InternVL3-8B 40.6% No training, No tools Apr 2025 Official leaderboard
5 GPT-5 mini 40.4% No training, Tools Aug 2025 Official leaderboard
6 VideoITG open weights 39.6% No training, No tools Official leaderboard
7 Gemini 3 Flash 37.6% No training, No tools Dec 2025 Official leaderboard
8 CoF open weights 35.9% No training, No tools Official leaderboard
9 Qwen3-VL-8B (GRPO†) open weights 35.6% Trained, Tools Official leaderboard
10 VideoLLAMA3-7B 31.6% No training, No tools Jan 2024 Official leaderboard
11 Grok 4.1 Fast 30.6% reasoning, No training, No tools Nov 2025 Official leaderboard
12 MaxInfo open weights 30.4% No training, No tools Official leaderboard
13 InternVL2-8B (PVC) open weights 30.2% No training, No tools Official leaderboard
14 AKS open weights 28.7% No training, No tools Official leaderboard
15 LLaVA-NeXT-Video-7B open weights 25.6% No training, No tools Official leaderboard
16 LLaVA-1.5-7B (VisionZip) open weights 23.8% No training, No tools Official leaderboard
17 Video-LLaVA-7B 21.9% No training, No tools Nov 2023 Official leaderboard
18 Qwen2.5-VL-7B (VisionZip) open weights 20.4% No training, No tools Official leaderboard
#ModelScoreSettingDateSource
1 Qwen3-VL-8B-Instruct 43.7% No training, No tools Oct 2025 Official leaderboard
2 InternVL3-8B 40.6% No training, No tools Apr 2025 Official leaderboard
3 GPT-5 mini 40.4% No training, Tools Aug 2025 Official leaderboard
4 Gemini 3 Flash 37.6% No training, No tools Dec 2025 Official leaderboard
5 VideoLLAMA3-7B 31.6% No training, No tools Jan 2024 Official leaderboard
6 Grok 4.1 Fast 30.6% reasoning, No training, No tools Nov 2025 Official leaderboard
7 Video-LLaVA-7B 21.9% No training, No tools Nov 2023 Official leaderboard

Rankings by part

We read 5 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Semantic Reasoning, Contextual Reasoning and Spatial Reasoning. Qwen3-VL-8B-Instruct leads 2 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.

Semantic Reasoning

7 models we track have a score here.

  1. 1 GPT-5 mini ↑2 58.4%
  2. 2 Qwen3-VL-8B-Instruct ↓1 58.2%
  3. 3 InternVL3-8B ↓1 53.4%
  4. 4 Gemini 3 Flash 48.2%
  5. 5 Grok 4.1 Fast ↑1 41.2%
  6. 6 VideoLLAMA3-7B ↓1 41%
  7. 7 Video-LLaVA-7B 24.4%

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
GPT-5 mini
No training… No training…
40.4%
Gemini 3 Flash
No training…
37.6%
Grok 4.1 Fast
reasoning, … reasoning, …
30.6%

Source: Official MarineEVT leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Qwen3-VL-8B-Instruct No training, No tools 43.7%
InternVL3-8B No training, No tools 40.6%
GPT-5 mini No training, Tools 40.4%
GPT-5 mini No training, No tools 37.7%
Gemini 3 Flash No training, No tools 37.6%
Gemini 3 Flash No training, Tools 36.3%
VideoLLAMA3-7B No training, No tools 31.6%
Grok 4.1 Fast reasoning, No training, No tools 30.6%
Grok 4.1 Fast reasoning, No training, Tools 25.1%
Video-LLaVA-7B No training, No tools 21.9%

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
GPT-5 mini
No training… No training…
40.4%
Gemini 3 Flash
No training…
37.6%
Grok 4.1 Fast
reasoning, … reasoning, …
30.6%

Source: Official MarineEVT leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Qwen3-VL-8B-Instruct No training, No tools 43.7%
InternVL3-8B No training, No tools 40.6%
GPT-5 mini No training, Tools 40.4%
GPT-5 mini No training, No tools 37.7%
Gemini 3 Flash No training, No tools 37.6%
Gemini 3 Flash No training, Tools 36.3%
VideoLLAMA3-7B No training, No tools 31.6%
Grok 4.1 Fast reasoning, No training, No tools 30.6%
Grok 4.1 Fast reasoning, No training, Tools 25.1%
Video-LLaVA-7B No training, No tools 21.9%

Number one over time

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

Qwen
Own leaderboard Not tested since Jun 2026

Source: Official MarineEVT 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 2026 Near the ceiling (85%) Qwen3-VL-8B-Instruct InternVL3-8B VideoLLAMA3-7B Video-LLaVA-7B 0% 20% 40% 60% 80% 100% 2024 2025 2026 Near the ceiling (85%) Qwen3-VL-8B-Instruct VideoLLAMA3-7B

Source: Official MarineEVT 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 Audio and video

All 695 in Audio and video
  • Video-MMMU In use 300 expert-level lecture videos with questions testing perception, comprehension and adaptation of knowledge. Accuracy. Qwen 3.8 Max 88.7%
  • LVBench In use Hours-long video understanding with multiple-choice questions on reasoning, grounding and summarization. Gemini 4 Argon 91.7%
  • MMVU In use 3,000 expert-annotated questions on videos across 27 disciplines. Multiple-choice and open-ended accuracy. Qwen 3.8 Max 82.4%
  • MLVU In use Multi-task long video understanding (minutes to hours) with multiple-choice and generation tasks. Accuracy. Qwen 3.8 Max 90.8%
  • FLEURS In use Models transcribe read speech in 102 languages built from FLoRes-101 sentences. Model cards usually report word error rate, so lower scores are better. Phi-4-multimodal 4.00%
  • VoiceBench In use Spoken instructions with speaker, environment and content variation. Voice assistant responses scored by GPT judge or rules. Qwen3.5-Omni-Plus 93.1%

Data audit

9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 11 months ago, in December 2025, for Gemini 3 Flash. 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 (18 entries).
  • Pass Models scored 7 models we track, plus 11 other entries on its leaderboard.
  • Warning Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Gemini 3 Flash (#60).
  • Pass Headroom The best is 43.7%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on marineevt.hkustvgd.com: 10 rows for models we track and 11 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