Benchmark Almanac

EgoProceVQA

Qwen3.5-Plus (Commercial) 65% tops its leaderboard GPT-5.1 51.6% leads the models we track
Not in model cards

Models watch first-person clips and answer questions about the steps in a procedure. Some questions ask when a step occurred or which step is missing.

The authors compare 13 models and one agent on 3,600 questions. The average excludes the task that asks for start and end times.

Released
July 2026
Kind
Video understanding
Level
Everyday
Scored
Checked automatically
Answers
Multiple choice
Measures
Capability
Language
English
In and out
Video, Text → Text
Interaction
Single turn
Items from
Generated
Refresh
One fixed set
Run by
Anyone
In views
Vision
Strongest model tested
Gemini 3 Flash #60 on the capability index
Last model added
December 2025
Scores refreshed
November 20th, 2026
14
Entries on its own leaderboard, 6 models we track

Qwen3.5-Plus (Commercial) tops the leaderboard with 65% GPT-5.1 leads with 51.6% 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 on its own with several settings, we keep its best. Every run is in the next chart. Agents other companies built around a model get a pale shade of that model's lab color. Gray marks blends and entries not in our model list, such as fine-tunes and older models. None of these count toward leaders. The best model we track is GPT-5.1, at 51.6%. Agents built around a model, such as Self-Skill-Exploration Agent + Qwen2.5-VL-7B, are their own entries. We don't count them as the model's score. 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 Qwen3.5-Plus (Commercial) 16 frames 65%
  2. 2 Self-Skill-Exploration Agent + Qwen2.5-VL-7B 16 frames 53%
  3. 3 GPT-5.1 8 frames 51.6%
  4. 4 InternVL3-38B 16 frames 51.3%
  5. 5 Gemini 3 Flash 8 frames 51.3%
  6. 6 GPT-4o Mar 2025, 8 frames 46.6%
  7. 7 Qwen2-VL-7B-Instruct 16 frames 43.9%
  8. 8 Qwen2.5-VL-7B 16 frames 42.9%
  9. 9 Qwen3-VL-8B 8 frames 41.9%
  10. 10 Video-LLaMA2 8 frames 39.4%
  11. 11 Vinci 8 frames 36.8%
  12. 12 LLaVA-OneVision 8 frames 36.4%
  13. 13 EgoGPT 8 frames 34.5%
  14. 14 Video-LLaVA 8 frames 30.4%

Source: Official EgoProceVQA leaderboard

#EntryScoreSettingDateSource
1 Qwen3.5-Plus (Commercial) 65% 16 frames Official leaderboard
2 Self-Skill-Exploration Agent + Qwen2.5-VL-7B 53% 16 frames Jan 2025 Official leaderboard
3 GPT-5.1 51.6% 8 frames Nov 2025 Official leaderboard
4 InternVL3-38B 51.3% 16 frames Apr 2025 Official leaderboard
5 Gemini 3 Flash 51.3% 8 frames Dec 2025 Official leaderboard
6 GPT-4o Mar 2025 46.6% 8 frames Mar 2025 Official leaderboard
7 Qwen2-VL-7B-Instruct 43.9% 16 frames Aug 2024 Official leaderboard
8 Qwen2.5-VL-7B 42.9% 16 frames Jan 2025 Official leaderboard
9 Qwen3-VL-8B 41.9% 8 frames Official leaderboard
10 Video-LLaMA2 39.4% 8 frames Official leaderboard
11 Vinci 36.8% 8 frames Official leaderboard
12 LLaVA-OneVision 36.4% 8 frames Official leaderboard
13 EgoGPT 34.5% 8 frames Official leaderboard
14 Video-LLaVA 30.4% 8 frames Official leaderboard
#ModelScoreSettingDateSource
1 GPT-5.1 51.6% 8 frames Nov 2025 Official leaderboard
2 InternVL3-38B 51.3% 16 frames Apr 2025 Official leaderboard
3 Gemini 3 Flash 51.3% 8 frames Dec 2025 Official leaderboard
4 GPT-4o Mar 2025 46.6% 8 frames Mar 2025 Official leaderboard
5 Qwen2-VL-7B-Instruct 43.9% 16 frames Aug 2024 Official leaderboard
6 Qwen2.5-VL-7B 42.9% 16 frames Jan 2025 Official leaderboard

Rankings by part

We read 10 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as KSR, PSR and KSG. InternVL3-38B leads 5 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

KSR

6 models we track have a score here.

  1. 1 GPT-5.1 66.5%
  2. 2 InternVL3-38B 62.8%
  3. 3 Gemini 3 Flash 62.7%
  4. 4 GPT-4o Mar 2025 59.3%
  5. 5 Qwen2-VL-7B-Instruct 57%
  6. 6 Qwen2.5-VL-7B 51.2%

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
Self-Skill-Exploration Agent + Qwen2.5-VL-7B
8 frames 16 frames
53%
InternVL3-38B
8 frames 16 frames
51.3%
Qwen2-VL-7B-Instruct
8 frames
43.9%
Qwen2.5-VL-7B
8 frames 16 frames
42.9%

Source: Official EgoProceVQA leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Self-Skill-Exploration Agent + Qwen2.5-VL-7B 16 frames Self-Skill-Exploration Agent 53%
GPT-5.1 8 frames 51.6%
InternVL3-38B 16 frames 51.3%
Gemini 3 Flash 8 frames 51.3%
Self-Skill-Exploration Agent + Qwen2.5-VL-7B 8 frames Self-Skill-Exploration Agent 49.2%
InternVL3-38B 8 frames 48.7%
GPT-4o Mar 2025 8 frames 46.6%
Qwen2-VL-7B-Instruct 16 frames 43.9%
Qwen2-VL-7B-Instruct 8 frames 43%
Qwen2.5-VL-7B 16 frames 42.9%
Qwen2.5-VL-7B 8 frames 40.3%

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
InternVL3-38B
8 frames 16 frames
51.3%
Qwen2-VL-7B-Instruct
8 frames 16 frames
43.9%
Qwen2.5-VL-7B
8 frames 16 frames
42.9%

Source: Official EgoProceVQA leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-5.1 8 frames 51.6%
InternVL3-38B 16 frames 51.3%
Gemini 3 Flash 8 frames 51.3%
InternVL3-38B 8 frames 48.7%
GPT-4o Mar 2025 8 frames 46.6%
Qwen2-VL-7B-Instruct 16 frames 43.9%
Qwen2-VL-7B-Instruct 8 frames 43%
Qwen2.5-VL-7B 16 frames 42.9%
Qwen2.5-VL-7B 8 frames 40.3%

Number one over time

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

OpenAI
Own leaderboard Not tested since Jul 2026
GPT-5.1 51.6%

Source: Official EgoProceVQA 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% 2025 2026 Near the ceiling (85%) GPT-5.1 InternVL3-38B GPT-4o Qwen2-VL-7B-Instruct 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) GPT-5.1 Qwen2-VL-7B-Instruct

Source: Official EgoProceVQA 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

8 of 10 checks pass, 2 warnings. Look first at the Task access check. We don’t know whether the tasks are public. Every benchmark

Catalog record 1 warning 1 pass
  • Warning Task access We don’t know whether the 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 (14 entries).
  • Pass Models scored 6 models we track, plus 8 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 51.6%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on z1oong.github.io: 11 rows for models we track and 7 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