CausalVQA
Cosmos3-Nano 70% 
Qwen2.5-VL-7B 49% leads since June 2025 Cosmos3-Nano 70% leads since May 2026 Models answer questions about real egocentric videos that probe physical cause and effect, in counterfactual, hypothetical, anticipation, planning and descriptive types.
Released with V-JEPA 2. Videos need an Ego4D license. The leaderboard scores a held-out split of 793 question pairs with private answers.
- Made by
- Meta FAIR
- Released
- June 2025
- Kind
- Physical prediction
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Multiple choice
- Tasks
- Public, plus a private test set
- Last in a system card
- May 2026
- Last model added
- May 2026
- Scores refreshed
- November 12th, 2026
Cosmos3-Nano leads with 70%
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
Qwen2.5-VL-7B leads with 49% across all independent runs
Best score per model across the official leaderboard, Epoch AI’s own runs and any board Epoch AI copied. Each bar and row names its source.
Cosmos3-Nano leads with 70%, claims included
Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model.
- 1 Cosmos3-Nano 70%
- 2 Cosmos3-Edge 37%
Source: Labs’ model cards
- 1 Qwen2.5-VL-7B Published baseline; vision ViT; LLM Qwen2.5-7B-Instruct 49%
- 2
InternVL2.5-8B Published baseline; vision InternViT-300M; LLM InternLM2.5-7B-Chat 47.5% - 3
LLaVA-OneVision 7B Published baseline; vision SigLIP; LLM Qwen2-7B 45.3% - 4
V-JEPA 2 Published baseline; vision VJEPA 2; LLM Llama3.1 8B 44.9%
Source: Official CausalVQA leaderboard
- 1 Cosmos3-Nano 70%
- 2 Qwen2.5-VL-7B Published baseline; vision ViT; LLM Qwen2.5-7B-Instruct 49%
- 3 InternVL2.5-8B Published baseline; vision InternViT-300M; LLM InternLM2.5-7B-Chat 47.5%
- 4 LLaVA-OneVision 7B Published baseline; vision SigLIP; LLM Qwen2-7B 45.3%
- 5 V-JEPA 2 Published baseline; vision VJEPA 2; LLM Llama3.1 8B 44.9%
- 6 Cosmos3-Edge 37%
Sources: Labs’ model cards and official CausalVQA leaderboard
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Cosmos3-Nano | 70% | May 2026 | Lab model card | |
| 2 | Cosmos3-Edge | 37% | May 2026 | Lab model card |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Qwen2.5-VL-7B | 49% | Published baseline; vision ViT; LLM Qwen2.5-7B-Instruct | Jan 2025 | Official leaderboard |
| 2 | InternVL2.5-8B | 47.5% | Published baseline; vision InternViT-300M; LLM InternLM2.5-7B-Chat | Dec 2024 | Official leaderboard |
| 3 | LLaVA-OneVision 7B | 45.3% | Published baseline; vision SigLIP; LLM Qwen2-7B | Aug 2024 | Official leaderboard |
| 4 | V-JEPA 2 | 44.9% | Published baseline; vision VJEPA 2; LLM Llama3.1 8B | Jun 2025 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Cosmos3-Nano | 70% | May 2026 | Lab model card | |
| 2 | Qwen2.5-VL-7B | 49% | Published baseline; vision ViT; LLM Qwen2.5-7B-Instruct | Jan 2025 | Official leaderboard |
| 3 | InternVL2.5-8B | 47.5% | Published baseline; vision InternViT-300M; LLM InternLM2.5-7B-Chat | Dec 2024 | Official leaderboard |
| 4 | LLaVA-OneVision 7B | 45.3% | Published baseline; vision SigLIP; LLM Qwen2-7B | Aug 2024 | Official leaderboard |
| 5 | V-JEPA 2 | 44.9% | Published baseline; vision VJEPA 2; LLM Llama3.1 8B | Jun 2025 | Official leaderboard |
| 6 | Cosmos3-Edge | 37% | May 2026 | Lab model card |
The benchmark's own leaderboard has 7 entries, 4 of them models we track. Switch the scores above to rank them together.
Number one over time Number one over time 2 models have held number one with claims included since its release in June 2025
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official CausalVQA leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Sources: Official CausalVQA leaderboard and labs’ model cards
Labs report it only in system cards
We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark. Some system cards do, below.
2 system cards report it, from 1 lab
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| May 31, 2026 | Cosmos3-Nano technical report | 70 | CausalVQA, Cosmos3-Nano 8B column, Table 10 | Capabilities |
| May 31, 2026 | Cosmos3-Edge technical report | 37 | CausalVQA, Cosmos3-Edge 2B column, Table 10 | Capabilities |
More in World models
- AgentWorldBench In use Language world models predict the next environment observation after an agent action, in 7 domains from MCP and search to Android and desktop OS. A judge model scores each prediction against the real observation on five dimensions. GPT-5.4 58.3
- IntPhys 2 In use Models watch simulated videos and judge whether each event is physically possible, testing object permanence, immutability, continuity and solidity. Scored by accuracy. Muse Spark Open 68.3%
- QwenWorldBench Fading Models act as world models that simulate agent environments in 7 domains, such as terminal, web, and Android. A rubric judge scores five dimensions. Qwen3.7-Plus 62.1%
- VPCT Fading Models view an image of a ball, ramps, and three buckets and predict which bucket the ball falls into. Scored by accuracy on 100 problems.
Gemini 3 Pro 91% - ContPhy Fading Models infer physical properties such as mass and density of fluids and soft bodies from videos and predict their dynamics. Seed2.0 Pro 67.4%
- RBench Only in system cards 650 image+text prompts for robot video generation (tasks and embodiments). MLLM judge scores task completion, physical plausibility, stability.
Wan 2.6 60.7%