Benchmark Almanac

Kinetics-700-2020

encord-team/ebind-audio-vision 65.5% tops its leaderboard Qwen3-VL-Embedding-8B 62% leads the models we track
Not in model cards
Version 700-2020

Models recognize human actions in short videos from the 2020 edition of Kinetics-700.

The authors replenished the clips for all 700 action categories.

Made by
DeepMind
Released
October 2020
Kind
Video understanding
Level
Everyday
Scored
Checked automatically
Answers
Prediction
Tasks
All public
Measures
Capability
Language
English
In and out
Video → Text
Interaction
Single turn
Items from
Another benchmark
Run by
Anyone
Based on
Kinetics-700
In views
Vision
Last model added
January 2026
Scores refreshed
November 20th, 2026
33
Entries on MTEB’s leaderboard, 3 models we track

encord-team/ebind-audio-vision tops the leaderboard with 65.5% Qwen3-VL-Embedding-8B leads with 62% among models released by January 2026

Best score per entry from MTEB’s leaderboard, read on November 20th, 2026. MTEB measures accuracy on its sample of the official test split, capped at about 16 clips per action class. 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-Embedding-8B, at 62%. This view shows only models we track, each at its best.

The newest model with a score here came out in January 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 encord-team/ebind-audio-vision Zero-shot video classification 65.5%
  2. 2 encord-team/ebind-full Zero-shot video classification 65.5%
  3. 3 encord-team/ebind-points-vision Zero-shot video classification 65.5%
  4. 4 facebook/pe-av-large Zero-shot video classification 64.8%
  5. 5 microsoft/xclip-large-patch14 Video classification 63.6%
  6. 6 LCO-Embedding/LCO-Embedding-Omni-7B Zero-shot video classification 63.1%
  7. 7 Qwen3-VL-Embedding-8B Zero-shot video classification 62%
  8. 8 facebook/pe-av-small Zero-shot video classification 61.6%
  9. 9 facebook/pe-av-base Zero-shot video classification 61.5%
  10. 10 Qwen3-VL-Embedding-2B Video classification 60.1%
  11. 11 LCO-Embedding/LCO-Embedding-Omni-3B Zero-shot video classification 59.8%
  12. 12 BidirLM/BidirLM-Omni-2.5B-Embedding Video and audio classification 58.1%
  13. 13 zhibinlan/UME-R1-7B Zero-shot video classification 57%
  14. 14 microsoft/xclip-base-patch16 Video classification 55.2%
  15. 15 zhibinlan/UME-R1-2B Zero-shot video classification 52.4%
  16. 16 Haon-Chen/e5-omni-7B Zero-shot video classification 52%
  17. 17 Tevatron/OmniEmbed-v0.1 Video classification 51.4%
  18. 18 microsoft/xclip-base-patch32 Video classification 48.8%
  19. 19 jinaai/jina-embeddings-v5-omni-small Video classification 46.8%
  20. 20 nvidia/omni-embed-nemotron-3b Video classification 46.5%
  21. 21 Haon-Chen/e5-omni-3B Zero-shot video classification 45.3%
  22. 22 VLM2Vec/VLM2Vec-V2.0 Video classification 44.4%
  23. 23 jinaai/jina-embeddings-v5-omni-nano Video classification 39%
  24. 24 facebook/vjepa2-vitg-fpc32-384-diving48 Video classification 23%
  25. 25 facebook/vjepa2-vitl-fpc16-256-ssv2 Video classification 22.9%
  26. 26 facebook/vjepa2-vitl-fpc32-256-diving48 Video classification 22.5%
  27. 27 facebook/vjepa2-vitl-fpc64-256 Video classification 22.4%
  28. 28 facebook/vjepa2-vitg-fpc64-384 Video classification 22%
  29. 29 facebook/vjepa2-vitg-fpc64-384-ssv2 Video classification 22%
  30. 30 facebook/vjepa2-vitg-fpc64-256 Video classification 21.9%
  31. 31 facebook/vjepa2-vith-fpc64-256 Video classification 17.9%
  32. 32 Qwen2.5-Omni-7B Video and audio classification 0.58%
  33. 33 Qwen/Qwen2.5-Omni-3B Video and audio classification 0.52%

Source: MTEB’s leaderboard

#EntryScoreSettingDateLinkSource
1 encord-team/ebind-audio-vision open weights, 0.8B 65.5% Zero-shot video classification huggingface.co Official leaderboard
2 encord-team/ebind-full open weights, 1.8B 65.5% Zero-shot video classification huggingface.co Official leaderboard
3 encord-team/ebind-points-vision open weights, 1.7B 65.5% Zero-shot video classification huggingface.co Official leaderboard
4 facebook/pe-av-large open weights, 2.2B 64.8% Zero-shot video classification huggingface.co Official leaderboard
5 microsoft/xclip-large-patch14 open weights, 0.6B 63.6% Video classification huggingface.co Official leaderboard
6 LCO-Embedding/LCO-Embedding-Omni-7B open weights, 8.9B 63.1% Zero-shot video classification huggingface.co Official leaderboard
7 Qwen3-VL-Embedding-8B 62% Zero-shot video classification Jan 2026 huggingface.co Official leaderboard
8 facebook/pe-av-small open weights, 0.8B 61.6% Zero-shot video classification huggingface.co Official leaderboard
9 facebook/pe-av-base open weights, 1B 61.5% Zero-shot video classification huggingface.co Official leaderboard
10 Qwen3-VL-Embedding-2B 60.1% Video classification Jan 2026 huggingface.co Official leaderboard
11 LCO-Embedding/LCO-Embedding-Omni-3B open weights, 4.7B 59.8% Zero-shot video classification huggingface.co Official leaderboard
12 BidirLM/BidirLM-Omni-2.5B-Embedding open weights, 2.4B 58.1% Video and audio classification huggingface.co Official leaderboard
13 zhibinlan/UME-R1-7B open weights, 8.3B 57% Zero-shot video classification huggingface.co Official leaderboard
14 microsoft/xclip-base-patch16 open weights, 0.2B 55.2% Video classification huggingface.co Official leaderboard
15 zhibinlan/UME-R1-2B open weights, 2.2B 52.4% Zero-shot video classification huggingface.co Official leaderboard
16 Haon-Chen/e5-omni-7B open weights, 8.9B 52% Zero-shot video classification huggingface.co Official leaderboard
17 Tevatron/OmniEmbed-v0.1 open weights, 8.9B 51.4% Video classification huggingface.co Official leaderboard
18 microsoft/xclip-base-patch32 open weights, 0.2B 48.8% Video classification huggingface.co Official leaderboard
19 jinaai/jina-embeddings-v5-omni-small open weights, 1.6B 46.8% Video classification huggingface.co Official leaderboard
20 nvidia/omni-embed-nemotron-3b open weights, 4.7B 46.5% Video classification huggingface.co Official leaderboard
21 Haon-Chen/e5-omni-3B open weights, 4.7B 45.3% Zero-shot video classification huggingface.co Official leaderboard
22 VLM2Vec/VLM2Vec-V2.0 open weights, 2.2B 44.4% Video classification huggingface.co Official leaderboard
23 jinaai/jina-embeddings-v5-omni-nano open weights, 1B 39% Video classification huggingface.co Official leaderboard
24 facebook/vjepa2-vitg-fpc32-384-diving48 open weights, 1.1B 23% Video classification huggingface.co Official leaderboard
25 facebook/vjepa2-vitl-fpc16-256-ssv2 open weights, 0.4B 22.9% Video classification huggingface.co Official leaderboard
26 facebook/vjepa2-vitl-fpc32-256-diving48 open weights, 0.4B 22.5% Video classification huggingface.co Official leaderboard
27 facebook/vjepa2-vitl-fpc64-256 open weights, 0.3B 22.4% Video classification huggingface.co Official leaderboard
28 facebook/vjepa2-vitg-fpc64-384 open weights, 1B 22% Video classification huggingface.co Official leaderboard
29 facebook/vjepa2-vitg-fpc64-384-ssv2 open weights, 1.1B 22% Video classification huggingface.co Official leaderboard
30 facebook/vjepa2-vitg-fpc64-256 open weights, 1B 21.9% Video classification huggingface.co Official leaderboard
31 facebook/vjepa2-vith-fpc64-256 open weights, 0.7B 17.9% Video classification huggingface.co Official leaderboard
32 Qwen2.5-Omni-7B 0.58% Video and audio classification Mar 2025 huggingface.co Official leaderboard
33 Qwen/Qwen2.5-Omni-3B open weights, 5.5B 0.52% Video and audio classification huggingface.co Official leaderboard
#ModelScoreSettingDateLinkSource
1 Qwen3-VL-Embedding-8B 62% Zero-shot video classification Jan 2026 huggingface.co Official leaderboard
2 Qwen3-VL-Embedding-2B 60.1% Video classification Jan 2026 huggingface.co Official leaderboard
3 Qwen2.5-Omni-7B 0.58% Video and audio classification Mar 2025 huggingface.co Official leaderboard

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
Qwen3-VL-Embedding-8B
Video class…
62%
Qwen3-VL-Embedding-2B
Zero-shot v…
60.1%
Qwen2.5-Omni-7B
Zero-shot v…
0.58%

Source: MTEB’s leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Qwen3-VL-Embedding-8B Zero-shot video classification 62%
Qwen3-VL-Embedding-8B Video classification 60.9%
Qwen3-VL-Embedding-2B Video classification 60.1%
Qwen3-VL-Embedding-2B Zero-shot video classification 58.4%
Qwen2.5-Omni-7B Video and audio classification 0.58%
Qwen2.5-Omni-7B Video classification 0.20%
Qwen2.5-Omni-7B Zero-shot video classification 0.16%
Qwen2.5-Omni-7B Zero-shot video and audio classification 0.16%

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
Qwen3-VL-Embedding-8B
Video class…
62%
Qwen3-VL-Embedding-2B
Zero-shot v…
60.1%
Qwen2.5-Omni-7B
Zero-shot v…
0.58%

Source: MTEB’s leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Qwen3-VL-Embedding-8B Zero-shot video classification 62%
Qwen3-VL-Embedding-8B Video classification 60.9%
Qwen3-VL-Embedding-2B Video classification 60.1%
Qwen3-VL-Embedding-2B Zero-shot video classification 58.4%
Qwen2.5-Omni-7B Video and audio classification 0.58%
Qwen2.5-Omni-7B Video classification 0.20%
Qwen2.5-Omni-7B Zero-shot video classification 0.16%
Qwen2.5-Omni-7B Zero-shot video and audio classification 0.16%

2 models have held number one on MTEB’s leaderboard since 2023

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

Qwen
MTEB leaderboard Not tested since Jan 2026

Source: MTEB’s 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 MTEB’s leaderboard Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Qwen3-VL-Embedding-8B Qwen2.5-Omni-7B 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Qwen3-VL-Embedding-8B Qwen2.5-Omni-7B

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

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 10 months ago, in January 2026, for Qwen3-VL-Embedding-2B. 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 (33 entries).
  • Pass Models scored 3 models we track, plus 30 other entries on its leaderboard.
  • Warning Newest score The newest came 10 months ago, in January 2026, for Qwen3-VL-Embedding-2B.
  • Note Frontier models None of its models are on the capabilities index.
  • Pass Headroom The best is 62%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on MTEB: 8 rows for models we track and 30 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