Kinetics-400
Version 400
Models recognize one of 400 human actions in short video clips.
MTEB reports accuracy on its Kinetics400VA task.
- Made by
- DeepMind
- Released
- May 2017
- Kind
- Video understanding
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Prediction
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Refresh
- New versions replace it
- Run by
- Anyone
- In views
- Vision, Audio and speech
- Last model added
- March 2025
- Scores refreshed
- November 20th, 2026
BidirLM/BidirLM-Omni-2.5B-Embedding tops the leaderboard with 67.6% Qwen2.5-Omni-7B leads with 0.25%
Best score per entry from MTEB’s leaderboard, read on November 20th, 2026. MTEB measures audio-video classification accuracy on its Kinetics400VA sample, with up to ten test clips per class. 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 Qwen2.5-Omni-7B, at 0.25%. This view shows only models we track, each at its best.
- 67.6 BidirLM/BidirLM-Omni-2.5B-Embedding
- 65.0 LCO-Embedding/LCO-Embedding-Omni-7B
- 63.3 LCO-Embedding/LCO-Embedding-Omni-3B
- 58.4 Haon-Chen/e5-omni-7B
- 56.5 nvidia/omni-embed-nemotron-3b
- 56.1 Tevatron/OmniEmbed-v0.1
- 55.2 jinaai/jina-embeddings-v5-omni-small
- 53.7 facebook/pe-av-base
- 53.2 facebook/pe-av-large
- 53.1 facebook/pe-av-small
- 46.5 Haon-Chen/e5-omni-3B
- 43.6 encord-team/ebind-audio-vision
- 43.6 encord-team/ebind-full
- 40.0 jinaai/jina-embeddings-v5-omni-nano
- 0.25 Qwen/Qwen2.5-Omni-3B
- 0.25
Qwen2.5-Omni-7B
- 1 BidirLM/BidirLM-Omni-2.5B-Embedding 67.6%
- 2 LCO-Embedding/LCO-Embedding-Omni-7B 65%
- 3 LCO-Embedding/LCO-Embedding-Omni-3B 63.3%
- 4 Haon-Chen/e5-omni-7B 58.4%
- 5 nvidia/omni-embed-nemotron-3b 56.5%
- 6 Tevatron/OmniEmbed-v0.1 56.1%
- 7 jinaai/jina-embeddings-v5-omni-small 55.2%
- 8 facebook/pe-av-base 53.7%
- 9 facebook/pe-av-large 53.2%
- 10 facebook/pe-av-small 53.1%
- 11 Haon-Chen/e5-omni-3B 46.5%
- 12 encord-team/ebind-audio-vision 43.6%
- 13 encord-team/ebind-full 43.6%
- 14 jinaai/jina-embeddings-v5-omni-nano 40%
- 15 Qwen/Qwen2.5-Omni-3B 0.25%
- 16
Qwen2.5-Omni-7B 0.25%
Source: MTEB’s leaderboard
- 1
Qwen2.5-Omni-7B 0.25%
Source: MTEB’s leaderboard
| # | Entry | Score | Setting | Date | Link | Source |
|---|---|---|---|---|---|---|
| 1 | BidirLM/BidirLM-Omni-2.5B-Embedding open weights, 2.4B | 67.6% | huggingface.co | Official leaderboard | ||
| 2 | LCO-Embedding/LCO-Embedding-Omni-7B open weights, 8.9B | 65% | huggingface.co | Official leaderboard | ||
| 3 | LCO-Embedding/LCO-Embedding-Omni-3B open weights, 4.7B | 63.3% | huggingface.co | Official leaderboard | ||
| 4 | Haon-Chen/e5-omni-7B open weights, 8.9B | 58.4% | huggingface.co | Official leaderboard | ||
| 5 | nvidia/omni-embed-nemotron-3b open weights, 4.7B | 56.5% | huggingface.co | Official leaderboard | ||
| 6 | Tevatron/OmniEmbed-v0.1 open weights, 8.9B | 56.1% | huggingface.co | Official leaderboard | ||
| 7 | jinaai/jina-embeddings-v5-omni-small open weights, 1.6B | 55.2% | huggingface.co | Official leaderboard | ||
| 8 | facebook/pe-av-base open weights, 1B | 53.7% | huggingface.co | Official leaderboard | ||
| 9 | facebook/pe-av-large open weights, 2.2B | 53.2% | huggingface.co | Official leaderboard | ||
| 10 | facebook/pe-av-small open weights, 0.8B | 53.1% | huggingface.co | Official leaderboard | ||
| 11 | Haon-Chen/e5-omni-3B open weights, 4.7B | 46.5% | huggingface.co | Official leaderboard | ||
| 12 | encord-team/ebind-audio-vision open weights, 0.8B | 43.6% | huggingface.co | Official leaderboard | ||
| 13 | encord-team/ebind-full open weights, 1.8B | 43.6% | huggingface.co | Official leaderboard | ||
| 14 | jinaai/jina-embeddings-v5-omni-nano open weights, 1B | 40% | huggingface.co | Official leaderboard | ||
| 15 | Qwen/Qwen2.5-Omni-3B open weights, 5.5B | 0.25% | huggingface.co | Official leaderboard | ||
| 16 | | 0.25% | Mar 2025 | huggingface.co | Official leaderboard |
| # | Model | Score | Setting | Date | Link | Source |
|---|---|---|---|---|---|---|
| 1 | | 0.25% | Mar 2025 | huggingface.co | Official leaderboard |
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: MTEB’s 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.
Versions
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
All 9 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (16 entries).
- Pass Models scored 1 model we track, plus 15 other entries on its leaderboard.
- Note Newest score The newest came 20 months ago, in March 2025, for Qwen2.5-Omni-7B. That’s expected for a test replaced by Kinetics-600.
- Note Frontier models None of its models are on the capabilities index.
- Pass Headroom The best is 0.25%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on MTEB: 1 row for models we track and 15 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.