Benchmark Almanac

TriWorldBench

cogito 72.9 tops its leaderboard
Not in model cards

World models generate synchronized robot-view videos from head and wrist cameras. Nineteen signals measure cross-view agreement, task alignment, physical coherence, motion, temporal stability, and visual quality.

The public validation set has 100 RoboTwin 2.0 episodes. Scoring combines automatic measurements with semantic model judgments.

Released
September 2026
Kind
Video world models
Level
Everyday
Scored
Graded by a model
Answers
Open-ended
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Images, Video → Video
Interaction
Single turn
Items from
Another benchmark
Refresh
One fixed set
Run by
Anyone
Based on
RoboTwin 2.0

In the authors’ words

Embodied world models predict the outcomes of robot actions to support learning and planning. For robots equipped with head and wrist cameras, this requires complementary views: the head view captures the overall task, while wrist views reveal local gripper-object interactions. However, evaluating these views independently cannot determine whether they describe the same action and object state. We introduce TRIWORLDBENCH, a benchmark for evaluating embodied world models through synchronized head, left-wrist, and right-wrist videos. It contains 500 episodes across 50 bimanual manipulation tasks and uses 19 metrics to assess tri-view consistency, task alignment, physical and 3D coherence, motion quality, temporal consistency, and visual quality. By combining cross-view checks with measurements tailored to each camera, the benchmark evaluates whether plausible individual videos also form a consistent prediction of the intended task. We summarize overall performance with TWB-Score and retain per-view results to identify where predictions fail. This extends world-model evaluation beyond single-view visual quality. Code, data, and metric definitions are available at https://github.com/TriWorldBench/TriWorldBench.

Abstract of TriWorldBench: A Tri-View Consistency Perspective on Embodied World Models (arXiv:2609.26314, CC0)
69
Entries on its own leaderboard, 0 models we track

cogito tops the leaderboard with 72.9

Best score per entry from the benchmark's own 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.

  1. 1 cogito 72.9
  2. 2 WorldIncept 71.8
  3. 3 Digger-v2 71.5
  4. 4 X-WM-Pro 70.9
  5. 5 salad 70.5
  6. 6 BWM-Pro 70.4
  7. 7 digger 70.2
  8. 8 Salad-0.4 70.2
  9. 9 Soil 68.3
  10. 10 X-Lab 67.9
  11. 11 LinearEWM 67.6
  12. 12 LeapBotWM 66.5
  13. 13 CasualWM 66.0
  14. 14 X-WM 65.7
  15. 15 dream4act 65.7
  16. 16 VisBot 65.7
  17. 17 BWM 65.5
  18. 18 WoVR_Plus 65.4
  19. 19 Sii 64.8
  20. 20 ZWM 64.6
  21. 21 WorldNative 64.4
  22. 22 HY_World 64.3
  23. 23 PhyxWM 64.3
  24. 24 t640_mix50 64.0
  25. 25 XiaomiAutoWM 63.9
  26. 26 xiao-world 63.8
  27. 27 BetaBWM 63.6
  28. 28 Fysiverse-Video 63.6
  29. 29 WMEpivoryn 62.1
  30. 30 CoWM 62.1
  31. 31 VWA-World 61.3
  32. 32 ROBOWM 60.8
  33. 33 TrivoryxWM 60.8
  34. 34 CogVerse 60.5
  35. 35 DynaWolrd-V2 60.4
  36. 36 Xi-WM-0.4 59.8
  37. 37 LibWM 59.3
  38. 38 ActoVision-X 58.8
  39. 39 TriCam-Dreamer 58.6
  40. 40 ViewWeaver 58.5
  41. 41 MirageFlow-V3 58.4
  42. 42 DynaWorld 58.4
  43. 43 VMA-World 58.2
  44. 44 RopeWM 57.7
  45. 45 Ra_1 57.3
  46. 46 KineWorld 56.4
  47. 47 SD2-WM 56.4
  48. 48 LightWM-Alpha 55.6
  49. 49 Kimo 55.0
  50. 50 KineJing-Pro 55.0
  51. 51 Flash-WM 54.5
  52. 52 CCoae 54.4
  53. 53 ArmW 53.9
  54. 54 WDK-XLAB 53.8
  55. 55 Ennerverse-AC 53.7
  56. 56 CWAM2 53.7
  57. 57 GentleWM 53.7
  58. 58 CWAM 53.5
  59. 59 CWAM0.5 53.4
  60. 60 DreamDojo 51.7
  61. 61 VMA_X-Era 51.4
  62. 62 Dream 49.9
  63. 63 WMxel 49.6
  64. 64 LatentGrip 49.2
  65. 65 Act-Cosmos 48.9
  66. 66 motus 42.4
  67. 67 KineJing 41.2
  68. 68 genie-envisioner 40.7
  69. 69 ctrl-world 39.0

Source: Official TriWorldBench leaderboard

#EntryScoreSettingDateSource
1 cogito 72.9 Nov 2026 Official leaderboard
2 WorldIncept 71.8 Oct 2026 Official leaderboard
3 Digger-v2 71.5 Oct 2026 Official leaderboard
4 X-WM-Pro 70.9 Oct 2026 Official leaderboard
5 salad 70.5 Oct 2026 Official leaderboard
6 BWM-Pro 70.4 Oct 2026 Official leaderboard
7 digger 70.2 Oct 2026 Official leaderboard
8 Salad-0.4 70.2 Oct 2026 Official leaderboard
9 Soil 68.3 Oct 2026 Official leaderboard
10 X-Lab 67.9 Nov 2026 Official leaderboard
11 LinearEWM 67.6 Nov 2026 Official leaderboard
12 LeapBotWM 66.5 Oct 2026 Official leaderboard
13 CasualWM 66.0 Oct 2026 Official leaderboard
14 X-WM 65.7 Oct 2026 Official leaderboard
15 dream4act 65.7 Oct 2026 Official leaderboard
16 VisBot 65.7 Oct 2026 Official leaderboard
17 BWM 65.5 Oct 2026 Official leaderboard
18 WoVR_Plus 65.4 Oct 2026 Official leaderboard
19 Sii 64.8 Oct 2026 Official leaderboard
20 ZWM 64.6 Oct 2026 Official leaderboard
21 WorldNative 64.4 Nov 2026 Official leaderboard
22 HY_World 64.3 Nov 2026 Official leaderboard
23 PhyxWM 64.3 Oct 2026 Official leaderboard
24 t640_mix50 64.0 Oct 2026 Official leaderboard
25 XiaomiAutoWM 63.9 Oct 2026 Official leaderboard
26 xiao-world 63.8 Oct 2026 Official leaderboard
27 BetaBWM 63.6 Oct 2026 Official leaderboard
28 Fysiverse-Video 63.6 Oct 2026 Official leaderboard
29 WMEpivoryn 62.1 Oct 2026 Official leaderboard
30 CoWM 62.1 Oct 2026 Official leaderboard
31 VWA-World 61.3 Oct 2026 Official leaderboard
32 ROBOWM 60.8 Oct 2026 Official leaderboard
33 TrivoryxWM 60.8 Oct 2026 Official leaderboard
34 CogVerse 60.5 Oct 2026 Official leaderboard
35 DynaWolrd-V2 60.4 Oct 2026 Official leaderboard
36 Xi-WM-0.4 59.8 Oct 2026 Official leaderboard
37 LibWM 59.3 Oct 2026 Official leaderboard
38 ActoVision-X 58.8 Oct 2026 Official leaderboard
39 TriCam-Dreamer 58.6 Oct 2026 Official leaderboard
40 ViewWeaver 58.5 Oct 2026 Official leaderboard
41 MirageFlow-V3 58.4 Oct 2026 Official leaderboard
42 DynaWorld 58.4 Oct 2026 Official leaderboard
43 VMA-World 58.2 Oct 2026 Official leaderboard
44 RopeWM 57.7 Oct 2026 Official leaderboard
45 Ra_1 57.3 Oct 2026 Official leaderboard
46 KineWorld 56.4 Oct 2026 Official leaderboard
47 SD2-WM 56.4 Oct 2026 Official leaderboard
48 LightWM-Alpha 55.6 Oct 2026 Official leaderboard
49 Kimo 55.0 Oct 2026 Official leaderboard
50 KineJing-Pro 55.0 Oct 2026 Official leaderboard
51 Flash-WM 54.5 Oct 2026 Official leaderboard
52 CCoae 54.4 Oct 2026 Official leaderboard
53 ArmW 53.9 Oct 2026 Official leaderboard
54 WDK-XLAB 53.8 Oct 2026 Official leaderboard
55 Ennerverse-AC 53.7 Oct 2026 Official leaderboard
56 CWAM2 53.7 Nov 2026 Official leaderboard
57 GentleWM 53.7 Oct 2026 Official leaderboard
58 CWAM 53.5 Oct 2026 Official leaderboard
59 CWAM0.5 53.4 Oct 2026 Official leaderboard
60 DreamDojo 51.7 Oct 2026 Official leaderboard
61 VMA_X-Era 51.4 Oct 2026 Official leaderboard
62 Dream 49.9 Oct 2026 Official leaderboard
63 WMxel 49.6 Oct 2026 Official leaderboard
64 LatentGrip 49.2 Oct 2026 Official leaderboard
65 Act-Cosmos 48.9 Oct 2026 Official leaderboard
66 motus 42.4 Oct 2026 Official leaderboard
67 KineJing 41.2 Oct 2026 Official leaderboard
68 genie-envisioner 40.7 Oct 2026 Official leaderboard
69 ctrl-world 39.0 Oct 2026 Official 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 World models

All 53 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%

Data audit

All 8 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 2 pass 3 notes
  • Pass Independent runs Scores from the official leaderboard (69 entries).
  • Pass Models scored 69 entries on its leaderboard, none a model we track.
  • Note Newest score No scores yet.
  • Note Frontier models No models scored yet.
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on triworldbench-triworldbench-space.hf.space: 69 entries at the last read, none a model we track.
  • 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