Benchmark Almanac

WorldScore

UniWorld-View 85.5% tops its leaderboard
Not in model cards

3,000 next-scene generation tasks with camera-trajectory layouts. Scores 3D/4D and video world generators on controllability, quality, dynamics.

Made by
Stanford
Released
April 2025
Kind
Video world models
Scored
Checked automatically
Answers
Open-ended
Tasks
All public

In the authors’ words

We introduce the WorldScore benchmark, the first unified benchmark for world generation. We decompose world generation into a sequence of next-scene generation tasks with explicit camera trajectory-based layout specifications, enabling unified evaluation of diverse approaches from 3D and 4D scene generation to video generation models. The WorldScore benchmark encompasses a curated dataset of 3,000 test examples that span diverse worlds: static and dynamic, indoor and outdoor, photorealistic and stylized. The WorldScore metrics evaluate generated worlds through three key aspects: controllability, quality, and dynamics. Through extensive evaluation of 19 representative models, including both open-source and closed-source ones, we reveal key insights and challenges for each category of models. Our dataset, evaluation code, and leaderboard can be found at https://haoyi-duan.github.io/WorldScore/

Abstract of WorldScore: A Unified Evaluation Benchmark for World Generation (arXiv:2504.00983, CC0)
35
Entries on its own leaderboard, 0 models we track

UniWorld-View tops the leaderboard with 85.5%

Best score per entry from the benchmark's own leaderboard, read on November 12th, 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.

35 entries. Scroll sideways for the rest.

  1. 85.5 UniWorld-View
  2. 85.1 WorldScape-0.2(MoE)
  3. 84.5 World Dreamer
  4. 83.5 EvoPhys-World
  5. 83.2 Inspatio-World
  6. 81.4 WorldScape-0.2
  7. 81.1 EonWorld
  8. 80.8 WorldScape-0.1
  9. 80.5 FantasyWorld-1.0
  10. 78.5 FantasyWorld-0.1
  11. 78.2 TeleWorld
  12. 77.6 Voyager
  13. 72.7 WonderWorld
  14. 70.8 FlashWorld
  15. 70.4 LucidDreamer
  16. 63.8 WonderJourney
  17. 62.1 CogVideoX-I2V
  18. 62.1 Text2Room
  19. 61.1 InvisibleStitch
  20. 60.7 Gen-3
  21. 57.6 Wan2.1
  22. 57.5 Hailuo
  23. 55.4 LTX-Video
  24. 55.3 Allegro
  25. 54.2 CogVideoX-T2V
  26. 52.9 EasyAnimate
  27. 52.6 VideoCrafter2
  28. 52.1 DynamiCrafter
  29. 50.7 SceneScape
  30. 50.5 VideoCrafter1-I2V
  31. 49.7 VWA-World
  32. 47.1 VideoCrafter1-T2V
  33. 45.6 T2V-Turbo
  34. 42.3 Vchitect-2.0
  35. 28.0 4D-fy
  1. 1 UniWorld-View 85.5%
  2. 2 WorldScape-0.2(MoE) 85.1%
  3. 3 World Dreamer 84.5%
  4. 4 EvoPhys-World 83.5%
  5. 5 Inspatio-World 83.2%
  6. 6 WorldScape-0.2 81.4%
  7. 7 EonWorld 81.1%
  8. 8 WorldScape-0.1 80.8%
  9. 9 FantasyWorld-1.0 80.5%
  10. 10 FantasyWorld-0.1 78.5%
  11. 11 TeleWorld 78.2%
  12. 12 Voyager 77.6%
  13. 13 WonderWorld 72.7%
  14. 14 FlashWorld 70.8%
  15. 15 LucidDreamer 70.4%
  16. 16 WonderJourney 63.8%
  17. 17 CogVideoX-I2V 62.1%
  18. 18 Text2Room 62.1%
  19. 19 InvisibleStitch 61.1%
  20. 20 Gen-3 60.7%
  21. 21 Wan2.1 57.6%
  22. 22 Hailuo 57.5%
  23. 23 LTX-Video 55.4%
  24. 24 Allegro 55.3%
  25. 25 CogVideoX-T2V 54.2%
  26. 26 EasyAnimate 52.9%
  27. 27 VideoCrafter2 52.6%
  28. 28 DynamiCrafter 52.1%
  29. 29 SceneScape 50.7%
  30. 30 VideoCrafter1-I2V 50.5%
  31. 31 VWA-World 49.7%
  32. 32 VideoCrafter1-T2V 47.1%
  33. 33 T2V-Turbo 45.6%
  34. 34 Vchitect-2.0 42.3%
  35. 35 4D-fy 28%

Source: Official WorldScore leaderboard

#EntryScoreSettingDateSource
1 UniWorld-View 85.5% Official leaderboard
2 WorldScape-0.2(MoE) 85.1% Official leaderboard
3 World Dreamer 84.5% Official leaderboard
4 EvoPhys-World 83.5% Official leaderboard
5 Inspatio-World 83.2% Official leaderboard
6 WorldScape-0.2 81.4% Official leaderboard
7 EonWorld 81.1% Official leaderboard
8 WorldScape-0.1 80.8% Official leaderboard
9 FantasyWorld-1.0 80.5% Official leaderboard
10 FantasyWorld-0.1 78.5% Official leaderboard
11 TeleWorld 78.2% Official leaderboard
12 Voyager 77.6% Official leaderboard
13 WonderWorld 72.7% Official leaderboard
14 FlashWorld 70.8% Official leaderboard
15 LucidDreamer 70.4% Official leaderboard
16 WonderJourney 63.8% Official leaderboard
17 CogVideoX-I2V 62.1% Official leaderboard
18 Text2Room 62.1% Official leaderboard
19 InvisibleStitch 61.1% Official leaderboard
20 Gen-3 60.7% Official leaderboard
21 Wan2.1 57.6% Official leaderboard
22 Hailuo 57.5% Official leaderboard
23 LTX-Video 55.4% Official leaderboard
24 Allegro 55.3% Official leaderboard
25 CogVideoX-T2V 54.2% Official leaderboard
26 EasyAnimate 52.9% Official leaderboard
27 VideoCrafter2 52.6% Official leaderboard
28 DynamiCrafter 52.1% Official leaderboard
29 SceneScape 50.7% Official leaderboard
30 VideoCrafter1-I2V 50.5% Official leaderboard
31 VWA-World 49.7% Official leaderboard
32 VideoCrafter1-T2V 47.1% Official leaderboard
33 T2V-Turbo 45.6% Official leaderboard
34 Vchitect-2.0 42.3% Official leaderboard
35 4D-fy 28% Official leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

More in World models

All 39 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%