Something-Something V2
Models recognize actions involving people manipulating everyday objects. The benchmark distinguishes actions that differ in direction, ordering, or outcome.
- Made by
- TwentyBN
- Kind
- Video understanding
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Short answer
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Refresh
- New versions replace it
- Run by
- Anyone
- Based on
- A version of another benchmark
- In views
- Vision
- Last model added
- June 2025
- Scores refreshed
- November 20th, 2026
MOSS-L tops the leaderboard with 75.3% V-JEPA 2 leads with 74%
Best score per entry from SOTA2 Research’s leaderboard, read on November 20th, 2026. 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 V-JEPA 2, at 74%. This view shows only models we track, each at its best.
- 75.3 MOSS-L
- 75.2 Side4Video
- 74.4 MAWS
- 74.3 V-JEPA
- 74.2 OmniMAE
- 74.0 MAE
- 74.0
V-JEPA 2
- 73.8 MoTED
- 73.5 ATM
- 73.1 DiST
- 73.0 UniFormerV2
- 72.4 MOSS-B
- 72.3 ST-Adapter
- 72.2 DualPath
- 71.5 StructViT-B-4-1
- 71.4 OMNIVORE (Swin-B)
- 71.2 UniFormer-B
- 71.1 StructViT-B-4-2
- 70.9 COVER
- 70.6 AIM
- 70.5 MViTv2-B
- 69.6 Swin-B
- 69.6 VideoSwin-B
- 69.5 ORVIT
- 69.1 MML
- 69.1 OmniCLIP
- 69.0 StructViT-S-8-1
- 68.8 StructViT-S-4-1
- 68.7 MVIT-B-24
- 68.7 MViT-B-24, 32x3
- 68.3 StructViT-S-4-2
- 68.1 Mformer-L
- 68.1 MFormer-L
- 68.1 VIMPAC
- 67.7 MViT-B, 64x3
- 67.7 UniFormer-S
- 67.3 M2CLIP
- 67.1 Mformer-HR
- 66.7 EVL
- 66.0 RSANet
- 65.9 ViViT
- 65.9 ViViT-L
- 65.7 SELFYNet
- 65.4 X-ViT
- 65.3 TDN
- 64.7 MSNet
- 64.6 EVEREST
- 64.5 CT-Net
- 64.3 VideoMAE
- 63.4 TSM
- 63.0 SlowFast
- 62.5 TimeSformer-HR
- 62.4 TimeSformer
- 62.3 TimeSformer-L
- 55.8 ρBYOLρ=2
- 55.0 RSPNet
- 54.4 ρMoCoρ=2
- 51.7 ρSwAVρ=2
- 48.8 TRN
- 48.7 Vita-CLIP B/16
- 38.1 B2
- 1 MOSS-L backbone=ViT-L/14, pre-train=Merged-2B, input=16×6, TFLOPs=1.67×6, protocol=Frozen backbone 75.3%
- 2 Side4Video backbone=ViT-E/14, pre-train=Merged-2B, input=16×6, TFLOPs=15.96×6, protocol=Frozen backbone 75.2%
- 3 MAWS Pre-training Dataset=IG-3B, Arch.=ViT-L, Res.=224 74.4%
- 4 V-JEPA backbone=ViT-H/16, pre-train=VM2M, input=16×6, protocol=Frozen backbone 74.3%
- 5 OmniMAE Pre-training Dataset=IN1k + SSv2, Arch.=ViT-L, Res.=224 74.2%
- 6 MAE Pre-training Dataset=K400, Arch.=ViT-L, Res.=224 74%
- 7
V-JEPA 2 backbone=ViT-H/16, pre-train=VM22M, input=16×6, protocol=Frozen backbone 74%
- 8 MoTED backbone=ViT-L/14, pre-train=CLIP400M, input=32×3, TFLOPs=2.89×3, protocol=Frozen backbone 73.8%
- 9 ATM backbone=ViT-L/14, pre-train=CLIP400M, input=16×6, TFLOPs=0.84×6, protocol=Full finetuning 73.5%
- 10 DiST backbone=ViT-L/14, pre-train=CLIP400M, input=32×3, TFLOPs=2.83×3, protocol=Frozen backbone 73.1%
- 11 UniFormerV2 backbone=ViT-L/14, pre-train=CLIP400M, input=32×3, TFLOPs=1.73×3, protocol=Full finetuning 73%
- 12 MOSS-B backbone=ViT-B/16, pre-train=CLIP400M, input=16×6, TFLOPs=0.36×6, protocol=Frozen backbone 72.4%
- 13 ST-Adapter backbone=ViT-L/14, pre-train=CLIP400M, input=32×3, TFLOPs=2.75×3, protocol=Frozen backbone 72.3%
- 14 DualPath backbone=ViT-L/14, pre-train=CLIP400M, input=48×3, TFLOPs=0.72×3, protocol=Frozen backbone 72.2%
- 15 StructViT-B-4-1 pretrain=K400, frame x crop x clip=32x3x1, FLOPs (G)=1963 71.5%
- 16 OMNIVORE (Swin-B) Modality=RGB-only 71.4%
- 17 UniFormer-B pretrain=K400, frame x crop x clip=32x3x1, FLOPs (G)=777 71.2%
- 18 StructViT-B-4-2 pretrain=K400, frame x crop x clip=32x3x1, FLOPs (G)=784 71.1%
- 19 COVER Modality=RGB-only 70.9%
- 20 AIM backbone=ViT-L/14, pre-train=CLIP400M, input=32×3, TFLOPs=3.84×3, protocol=Frozen backbone 70.6%
- 21 MViTv2-B pretrain=K400, frame x crop x clip=32x3x1, FLOPs (G)=675 70.5%
- 22 Swin-B Backbone protocol=Finetuned, Zero-shot capability=false 69.6%
- 23 VideoSwin-B Modality=RGB-only 69.6%
- 24 ORVIT Modality=RGB-only 69.5%
- 25 MML Modality=Multi-modal 69.1%
- 26 OmniCLIP backbone=ViT-B/16, pre-train=CLIP400M, input=32×3, TFLOPs=0.8×3, protocol=Frozen backbone 69.1%
- 27 StructViT-S-8-1 pretrain=K400, frame x crop x clip=16x3x1, FLOPs (G)=405 69%
- 28 StructViT-S-4-1 pretrain=K400, frame x crop x clip=16x3x1, FLOPs (G)=246 68.8%
- 29 MVIT-B-24 Modality=RGB-only 68.7%
- 30 MViT-B-24, 32x3 pretrain=K600, frame x crop x clip=32x1x3, FLOPs (G)=708 68.7%
- 31 StructViT-S-4-2 pretrain=K400, frame x crop x clip=16x3x1, FLOPs (G)=126 68.3%
- 32 Mformer-L pretrain=K400, frame x crop x clip=32x3x1, FLOPs (G)=3555 68.1%
- 33 MFormer-L Modality=RGB-only 68.1%
- 34 VIMPAC Modality=RGB-only 68.1%
- 35 MViT-B, 64x3 pretrain=K400, frame x crop x clip=64x1x3, FLOPs (G)=1365 67.7%
- 36 UniFormer-S pretrain=K400, frame x crop x clip=16x3x1, FLOPs (G)=125 67.7%
- 37 M2CLIP backbone=ViT-B/16, pre-train=CLIP400M, input=16×12, protocol=Frozen backbone 67.3%
- 38 Mformer-HR pretrain=K400, frame x crop x clip=16x3x1, FLOPs (G)=2876 67.1%
- 39 EVL backbone=ViT-L/14, pre-train=CLIP400M, input=32×3, TFLOPs=3.21×3, protocol=Frozen backbone 66.7%
- 40 RSANet pretrain=IN-1K, frame x crop x clip=16x1x1, FLOPs (G)=72 66%
- 41 ViViT Backbone protocol=Finetuned, Zero-shot capability=false 65.9%
- 42 ViViT-L Modality=RGB-only 65.9%
- 43 SELFYNet pretrain=IN-1K, frame x crop x clip=16x1x1, FLOPs (G)=77 65.7%
- 44 X-ViT pretrain=IN-21K, frame x crop x clip=32x3x1, FLOPs (G)=1270 65.4%
- 45 TDN pretrain=IN-1K, frame x crop x clip=16x1x1, FLOPs (G)=72 65.3%
- 46 MSNet pretrain=IN-1K, frame x crop x clip=16x1x1, FLOPs (G)=101 64.7%
- 47 EVEREST Backbone=ViT-B, Extra data=None 64.6%
- 48 CT-Net pretrain=IN-1K, frame x crop x clip=16x1x1, FLOPs (G)=75 64.5%
- 49 VideoMAE Backbone=ViT-B, Extra data=None 64.3%
- 50 TSM Backbone protocol=Finetuned, Zero-shot capability=false 63.4%
- 51 SlowFast Modality=RGB-only 63%
- 52 TimeSformer-HR pretrain=IN-21K, frame x crop x clip=96x3x1, FLOPs (G)=5109 62.5%
- 53 TimeSformer Modality=RGB-only 62.4%
- 54 TimeSformer-L pretrain=IN-21K, frame x crop x clip=16x3x1, FLOPs (G)=7140 62.3%
- 55 ρBYOLρ=2 Backbone=SR50, Extra data=K400 55.8%
- 56 RSPNet Backbone=S3D-G, Extra data=K400 55%
- 57 ρMoCoρ=2 Backbone=SR50, Extra data=K400 54.4%
- 58 ρSwAVρ=2 Backbone=SR50, Extra data=K400 51.7%
- 59 TRN Backbone protocol=Finetuned, Zero-shot capability=false 48.8%
- 60 Vita-CLIP B/16 Backbone protocol=Frozen, Zero-shot capability=true, Mc=8, Ms=8 48.7%
- 61 B2 Backbone protocol=Frozen, Zero-shot capability=true 38.1%
Source: SOTA2 Research’s leaderboard
Source: SOTA2 Research’s leaderboard
| # | Entry | Score | Setting | Date | Link | Source |
|---|---|---|---|---|---|---|
| 1 | MOSS-L | 75.3% | backbone=ViT-L/14, pre-train=Merged-2B, input=16×6, TFLOPs=1.67×6, protocol=Frozen backbone | sota2.com | Official leaderboard | |
| 2 | Side4Video | 75.2% | backbone=ViT-E/14, pre-train=Merged-2B, input=16×6, TFLOPs=15.96×6, protocol=Frozen backbone | sota2.com | Official leaderboard | |
| 3 | MAWS | 74.4% | Pre-training Dataset=IG-3B, Arch.=ViT-L, Res.=224 | sota2.com | Official leaderboard | |
| 4 | V-JEPA | 74.3% | backbone=ViT-H/16, pre-train=VM2M, input=16×6, protocol=Frozen backbone | sota2.com | Official leaderboard | |
| 5 | OmniMAE | 74.2% | Pre-training Dataset=IN1k + SSv2, Arch.=ViT-L, Res.=224 | sota2.com | Official leaderboard | |
| 6 | MAE | 74% | Pre-training Dataset=K400, Arch.=ViT-L, Res.=224 | sota2.com | Official leaderboard | |
| 7 | | 74% | backbone=ViT-H/16, pre-train=VM22M, input=16×6, protocol=Frozen backbone | Jun 2025 | sota2.com | Official leaderboard |
| 8 | MoTED | 73.8% | backbone=ViT-L/14, pre-train=CLIP400M, input=32×3, TFLOPs=2.89×3, protocol=Frozen backbone | sota2.com | Official leaderboard | |
| 9 | ATM | 73.5% | backbone=ViT-L/14, pre-train=CLIP400M, input=16×6, TFLOPs=0.84×6, protocol=Full finetuning | sota2.com | Official leaderboard | |
| 10 | DiST | 73.1% | backbone=ViT-L/14, pre-train=CLIP400M, input=32×3, TFLOPs=2.83×3, protocol=Frozen backbone | sota2.com | Official leaderboard | |
| 11 | UniFormerV2 | 73% | backbone=ViT-L/14, pre-train=CLIP400M, input=32×3, TFLOPs=1.73×3, protocol=Full finetuning | sota2.com | Official leaderboard | |
| 12 | MOSS-B | 72.4% | backbone=ViT-B/16, pre-train=CLIP400M, input=16×6, TFLOPs=0.36×6, protocol=Frozen backbone | sota2.com | Official leaderboard | |
| 13 | ST-Adapter | 72.3% | backbone=ViT-L/14, pre-train=CLIP400M, input=32×3, TFLOPs=2.75×3, protocol=Frozen backbone | sota2.com | Official leaderboard | |
| 14 | DualPath | 72.2% | backbone=ViT-L/14, pre-train=CLIP400M, input=48×3, TFLOPs=0.72×3, protocol=Frozen backbone | sota2.com | Official leaderboard | |
| 15 | StructViT-B-4-1 | 71.5% | pretrain=K400, frame x crop x clip=32x3x1, FLOPs (G)=1963 | sota2.com | Official leaderboard | |
| 16 | OMNIVORE (Swin-B) | 71.4% | Modality=RGB-only | sota2.com | Official leaderboard | |
| 17 | UniFormer-B | 71.2% | pretrain=K400, frame x crop x clip=32x3x1, FLOPs (G)=777 | sota2.com | Official leaderboard | |
| 18 | StructViT-B-4-2 | 71.1% | pretrain=K400, frame x crop x clip=32x3x1, FLOPs (G)=784 | sota2.com | Official leaderboard | |
| 19 | COVER | 70.9% | Modality=RGB-only | sota2.com | Official leaderboard | |
| 20 | AIM | 70.6% | backbone=ViT-L/14, pre-train=CLIP400M, input=32×3, TFLOPs=3.84×3, protocol=Frozen backbone | sota2.com | Official leaderboard | |
| 21 | MViTv2-B | 70.5% | pretrain=K400, frame x crop x clip=32x3x1, FLOPs (G)=675 | sota2.com | Official leaderboard | |
| 22 | Swin-B | 69.6% | Backbone protocol=Finetuned, Zero-shot capability=false | sota2.com | Official leaderboard | |
| 23 | VideoSwin-B | 69.6% | Modality=RGB-only | sota2.com | Official leaderboard | |
| 24 | ORVIT | 69.5% | Modality=RGB-only | sota2.com | Official leaderboard | |
| 25 | MML | 69.1% | Modality=Multi-modal | sota2.com | Official leaderboard | |
| 26 | OmniCLIP | 69.1% | backbone=ViT-B/16, pre-train=CLIP400M, input=32×3, TFLOPs=0.8×3, protocol=Frozen backbone | sota2.com | Official leaderboard | |
| 27 | StructViT-S-8-1 | 69% | pretrain=K400, frame x crop x clip=16x3x1, FLOPs (G)=405 | sota2.com | Official leaderboard | |
| 28 | StructViT-S-4-1 | 68.8% | pretrain=K400, frame x crop x clip=16x3x1, FLOPs (G)=246 | sota2.com | Official leaderboard | |
| 29 | MVIT-B-24 | 68.7% | Modality=RGB-only | sota2.com | Official leaderboard | |
| 30 | MViT-B-24, 32x3 | 68.7% | pretrain=K600, frame x crop x clip=32x1x3, FLOPs (G)=708 | sota2.com | Official leaderboard | |
| 31 | StructViT-S-4-2 | 68.3% | pretrain=K400, frame x crop x clip=16x3x1, FLOPs (G)=126 | sota2.com | Official leaderboard | |
| 32 | Mformer-L | 68.1% | pretrain=K400, frame x crop x clip=32x3x1, FLOPs (G)=3555 | sota2.com | Official leaderboard | |
| 33 | MFormer-L | 68.1% | Modality=RGB-only | sota2.com | Official leaderboard | |
| 34 | VIMPAC | 68.1% | Modality=RGB-only | sota2.com | Official leaderboard | |
| 35 | MViT-B, 64x3 | 67.7% | pretrain=K400, frame x crop x clip=64x1x3, FLOPs (G)=1365 | sota2.com | Official leaderboard | |
| 36 | UniFormer-S | 67.7% | pretrain=K400, frame x crop x clip=16x3x1, FLOPs (G)=125 | sota2.com | Official leaderboard | |
| 37 | M2CLIP | 67.3% | backbone=ViT-B/16, pre-train=CLIP400M, input=16×12, protocol=Frozen backbone | sota2.com | Official leaderboard | |
| 38 | Mformer-HR | 67.1% | pretrain=K400, frame x crop x clip=16x3x1, FLOPs (G)=2876 | sota2.com | Official leaderboard | |
| 39 | EVL | 66.7% | backbone=ViT-L/14, pre-train=CLIP400M, input=32×3, TFLOPs=3.21×3, protocol=Frozen backbone | sota2.com | Official leaderboard | |
| 40 | RSANet | 66% | pretrain=IN-1K, frame x crop x clip=16x1x1, FLOPs (G)=72 | sota2.com | Official leaderboard | |
| 41 | ViViT | 65.9% | Backbone protocol=Finetuned, Zero-shot capability=false | sota2.com | Official leaderboard | |
| 42 | ViViT-L | 65.9% | Modality=RGB-only | sota2.com | Official leaderboard | |
| 43 | SELFYNet | 65.7% | pretrain=IN-1K, frame x crop x clip=16x1x1, FLOPs (G)=77 | sota2.com | Official leaderboard | |
| 44 | X-ViT | 65.4% | pretrain=IN-21K, frame x crop x clip=32x3x1, FLOPs (G)=1270 | sota2.com | Official leaderboard | |
| 45 | TDN | 65.3% | pretrain=IN-1K, frame x crop x clip=16x1x1, FLOPs (G)=72 | sota2.com | Official leaderboard | |
| 46 | MSNet | 64.7% | pretrain=IN-1K, frame x crop x clip=16x1x1, FLOPs (G)=101 | sota2.com | Official leaderboard | |
| 47 | EVEREST | 64.6% | Backbone=ViT-B, Extra data=None | sota2.com | Official leaderboard | |
| 48 | CT-Net | 64.5% | pretrain=IN-1K, frame x crop x clip=16x1x1, FLOPs (G)=75 | sota2.com | Official leaderboard | |
| 49 | VideoMAE | 64.3% | Backbone=ViT-B, Extra data=None | sota2.com | Official leaderboard | |
| 50 | TSM | 63.4% | Backbone protocol=Finetuned, Zero-shot capability=false | sota2.com | Official leaderboard | |
| 51 | SlowFast | 63% | Modality=RGB-only | sota2.com | Official leaderboard | |
| 52 | TimeSformer-HR | 62.5% | pretrain=IN-21K, frame x crop x clip=96x3x1, FLOPs (G)=5109 | sota2.com | Official leaderboard | |
| 53 | TimeSformer | 62.4% | Modality=RGB-only | sota2.com | Official leaderboard | |
| 54 | TimeSformer-L | 62.3% | pretrain=IN-21K, frame x crop x clip=16x3x1, FLOPs (G)=7140 | sota2.com | Official leaderboard | |
| 55 | ρBYOLρ=2 | 55.8% | Backbone=SR50, Extra data=K400 | sota2.com | Official leaderboard | |
| 56 | RSPNet | 55% | Backbone=S3D-G, Extra data=K400 | sota2.com | Official leaderboard | |
| 57 | ρMoCoρ=2 | 54.4% | Backbone=SR50, Extra data=K400 | sota2.com | Official leaderboard | |
| 58 | ρSwAVρ=2 | 51.7% | Backbone=SR50, Extra data=K400 | sota2.com | Official leaderboard | |
| 59 | TRN | 48.8% | Backbone protocol=Finetuned, Zero-shot capability=false | sota2.com | Official leaderboard | |
| 60 | Vita-CLIP B/16 | 48.7% | Backbone protocol=Frozen, Zero-shot capability=true, Mc=8, Ms=8 | sota2.com | Official leaderboard | |
| 61 | B2 | 38.1% | Backbone protocol=Frozen, Zero-shot capability=true | sota2.com | 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: SOTA2 Research’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.
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
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 17 months ago, in June 2025, for V-JEPA 2. 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 (61 entries).
- Pass Models scored 1 model we track, plus 60 other entries on its leaderboard.
- Warning Newest score The newest came 17 months ago, in June 2025, for V-JEPA 2.
- Note Frontier models None of its models are on the capabilities index.
- Pass Headroom The best is 74%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on SOTA2 Research: 1 row for models we track and 60 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.