Mind the Gap
Models describe visible events during natural pauses in a video's soundtrack. The evaluator renders the narration and checks its timing and coverage.
- Made by
- Google Research and UC Berkeley
- Released
- October 2026
- Kind
- Video captioning
- Level
- University
- Scored
- Graded by a model
- Answers
- Open-ended
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Vision, Audio and speech
- Strongest model tested
-
Gemini 3.5 Flash #44 on the capability index
- Last model added
- May 2026
- Scores refreshed
- November 20th, 2026
Qwen3.5-Omni-Flash leads with 0.94 among models released by May 2026 Qwen3.5-Omni-Flash leads with 0.94 among models released by May 2026
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The paper gives no overall score. Prioritization measures how well descriptions cover salient visual events. 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. This view shows only models we track, each at its best.
The newest model with a score here came out in May 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 1
Qwen3.5-Omni-Flash Guideline and timing ASR, Gemini-3.5 Flash judge 0.94
- 2
Gemini 3.5 Flash Guideline and timing ASR, Gemini-3.5 Flash judge 0.93
- 3
Qwen3.5-Omni-Plus Guideline and timing ASR, Gemini-3.5 Flash judge 0.79
- 4
Gemini 2.5 Flash Sep 2025, Guideline and timing ASR, Gemini-2.5 Flash judge 0.76
- 5 Claude 4.5 Guideline and timing ASR, Gemini-2.5 Flash judge 0.59
- 6 Qwen-2-VL-Max Guideline and timing ASR, Gemini-2.5 Flash judge 0.59
- 7 GPT-5.2 Guideline and timing ASR, Gemini-2.5 Flash judge 0.55
- 8 GPT-4.1 mini Guideline and timing ASR, Gemini-2.5 Flash judge 0.52
- 9 Claude 4 Guideline and timing ASR, Gemini-2.5 Flash judge 0.51
Source: The Mind the Gap paper
- 1
Qwen3.5-Omni-Flash Guideline and timing ASR, Gemini-3.5 Flash judge 0.94
- 2
Gemini 3.5 Flash Guideline and timing ASR, Gemini-3.5 Flash judge 0.93
- 3
Qwen3.5-Omni-Plus Guideline and timing ASR, Gemini-3.5 Flash judge 0.79
- 4
Gemini 2.5 Flash Sep 2025, Guideline and timing ASR, Gemini-2.5 Flash judge 0.76
- 5 GPT-5.2 Guideline and timing ASR, Gemini-2.5 Flash judge 0.55
- 6 GPT-4.1 mini Guideline and timing ASR, Gemini-2.5 Flash judge 0.52
Source: The Mind the Gap paper
| # | Entry | Score ± | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 0.94 ±0 | Guideline and timing ASR, Gemini-3.5 Flash judge | Mar 2026 | Official leaderboard |
| 2 | | 0.93 ±0 | Guideline and timing ASR, Gemini-3.5 Flash judge | May 2026 | Official leaderboard |
| 3 | | 0.79 ±0.1 | Guideline and timing ASR, Gemini-3.5 Flash judge | Mar 2026 | Official leaderboard |
| 4 | | 0.76 ±0 | Guideline and timing ASR, Gemini-2.5 Flash judge | Sep 2025 | Official leaderboard |
| 5 | Claude 4.5 | 0.59 ±0 | Guideline and timing ASR, Gemini-2.5 Flash judge | Official leaderboard | |
| 6 | Qwen-2-VL-Max | 0.59 ±0 | Guideline and timing ASR, Gemini-2.5 Flash judge | Official leaderboard | |
| 7 | GPT-5.2 | 0.55 ±0 | Guideline and timing ASR, Gemini-2.5 Flash judge | Dec 2025 | Official leaderboard |
| 8 | GPT-4.1 mini | 0.52 ±0 | Guideline and timing ASR, Gemini-2.5 Flash judge | Apr 2025 | Official leaderboard |
| 9 | Claude 4 | 0.51 ±0.1 | Guideline and timing ASR, Gemini-2.5 Flash judge | Official leaderboard |
| # | Model | Score ± | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 0.94 ±0 | Guideline and timing ASR, Gemini-3.5 Flash judge | Mar 2026 | Official leaderboard |
| 2 | | 0.93 ±0 | Guideline and timing ASR, Gemini-3.5 Flash judge | May 2026 | Official leaderboard |
| 3 | | 0.79 ±0.1 | Guideline and timing ASR, Gemini-3.5 Flash judge | Mar 2026 | Official leaderboard |
| 4 | | 0.76 ±0 | Guideline and timing ASR, Gemini-2.5 Flash judge | Sep 2025 | Official leaderboard |
| 5 | GPT-5.2 | 0.55 ±0 | Guideline and timing ASR, Gemini-2.5 Flash judge | Dec 2025 | Official leaderboard |
| 6 | GPT-4.1 mini | 0.52 ±0 | Guideline and timing ASR, Gemini-2.5 Flash judge | Apr 2025 | Official leaderboard |
Rankings by part
We read 10 parts from the benchmark’s paper, each ranking models on one part of the test, such as No Redundancy, On-Screen Text and Timeliness. Qwen3.5-Omni-Flash leads 4 of them, and 4 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
No Redundancy
6 models we track have a score here.
- 1
Gemini 3.5 Flash ↑1 0.90
- 2
Qwen3.5-Omni-Flash ↓1 0.88
- 3
Qwen3.5-Omni-Plus 0.88
- 4 GPT-5.2 ↑1 0.73
- 5 GPT-4.1 mini ↑1 0.70
- 6
Gemini 2.5 Flash Sep 2025 ↓2 0.70
On-Screen Text
6 models we track have a score here.
- 1
Gemini 2.5 Flash Sep 2025 ↑3 0.64
- 2 GPT-5.2 ↑3 0.63
- 3
Qwen3.5-Omni-Flash ↓2 0.61
- 4 GPT-4.1 mini ↑2 0.58
- 5
Qwen3.5-Omni-Plus ↓2 0.58
- 6
Gemini 3.5 Flash ↓4 0.57
Timeliness
Seconds. 6 models we track have a score here.
- 1
Qwen3.5-Omni-Plus ↑2 2.67
- 2
Gemini 3.5 Flash 2.99
- 3
Qwen3.5-Omni-Flash ↓2 4.91
- 4
Gemini 2.5 Flash Sep 2025 5.00
- 5 GPT-5.2 22.6
- 6 GPT-4.1 mini 65.3
Audio Overlaps
Seconds. 6 models we track have a score here.
- 1
Gemini 3.5 Flash ↑1 0.26
- 2
Gemini 2.5 Flash Sep 2025 ↑2 8.92
- 3 GPT-5.2 ↑2 11.1
- 4 GPT-4.1 mini ↑2 21.5
- 5
Qwen3.5-Omni-Plus ↓2 30.4
- 6
Qwen3.5-Omni-Flash ↓5 52.1
Audio Coverage
A ratio of one fills the available silence. 6 models we track have a score here.
- 1 GPT-4.1 mini ↑5 1.48
- 2
Qwen3.5-Omni-Flash ↓1 0.94
- 3
Gemini 2.5 Flash Sep 2025 ↑1 0.74
- 4
Qwen3.5-Omni-Plus ↓1 0.72
- 5 GPT-5.2 0.65
- 6
Gemini 3.5 Flash ↓4 0.45
Average Silence
Seconds. 6 models we track have a score here.
- 1
Gemini 3.5 Flash ↑1 4.85
- 2
Gemini 2.5 Flash Sep 2025 ↑2 6.63
- 3
Qwen3.5-Omni-Plus 9.49
- 4 GPT-5.2 ↑1 14.8
- 5
Qwen3.5-Omni-Flash ↓4 16.6
- 6 GPT-4.1 mini 19.4
VideoA11y-Ref Descriptive
6 models we track have a score here.
- 1
Qwen3.5-Omni-Flash 4.43
- 2
Gemini 3.5 Flash 4.02
- 3
Qwen3.5-Omni-Plus 3.82
- 4 GPT-4.1 mini ↑2 3.28
- 5
Gemini 2.5 Flash Sep 2025 ↓1 3.10
- 6 GPT-5.2 ↓1 2.89
VideoA11y-Ref Objective
6 models we track have a score here.
- 1
Qwen3.5-Omni-Flash 4.41
- 2
Gemini 3.5 Flash 4.07
- 3
Qwen3.5-Omni-Plus 4.03
- 4 GPT-5.2 ↑1 2.94
- 5
Gemini 2.5 Flash Sep 2025 ↓1 2.92
- 6 GPT-4.1 mini 2.82
VideoA11y-Ref Accurate
6 models we track have a score here.
- 1
Qwen3.5-Omni-Flash 3.98
- 2
Gemini 3.5 Flash 3.79
- 3
Qwen3.5-Omni-Plus 3.23
- 4
Gemini 2.5 Flash Sep 2025 1.63
- 5 GPT-5.2 1.35
- 6 GPT-4.1 mini 1.31
VideoA11y-Ref Clear
6 models we track have a score here.
- 1
Qwen3.5-Omni-Flash 4.27
- 2
Gemini 3.5 Flash 4.02
- 3
Qwen3.5-Omni-Plus 3.81
- 4
Gemini 2.5 Flash Sep 2025 2.69
- 5 GPT-5.2 2.54
- 6 GPT-4.1 mini 2.19
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The Mind the Gap paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official Mind the Gap 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
- 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 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 6 months ago, in May 2026, for Gemini 3.5 Flash. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
- Pass Independent runs Scores from the benchmark’s paper (9 entries).
- Pass Models scored 6 models we track, plus 3 other entries on its leaderboard.
- Warning Newest score The newest came 6 months ago, in May 2026, for Gemini 3.5 Flash.
- Note Frontier models None of the top 10 has a score. The strongest tested is Gemini 3.5 Flash (#44).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 6 rows for models we track and 3 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.