1X World Model eval

1X World Model evaluation (1XWM)

How to read this picture

1X's internal video world model that predicts humanoid task success to rank policy checkpoints before real trials.1

Sources
Last checked 10 Oct 2026Basic entry8 of 18 facts checked at the sourceNext check 8 Apr 2027
Runs in
AI simulator1
Checked against real robots
Not checked
Skill
World models
Robot
Humanoid1

Comparisons with real robots

Tried on real robots, not compared1

Details

Self-measured by 1X. Checkpoints from 2 policy training runs and several architecture variants were scored in the world model and in double-blind real A/B runs on the Arcade task; plots show agreement but no correlation statistic. Success-prediction 'alignment' 63.06% (Shelf only) -> 71.17% (with Arcade data). Analytic claim: with 70% alignment and a 15% true gap, picks the better policy 90% of the time.

Known problems 1

  1. Struggles with unseen objects; pose errors accumulate in locomotion, limiting long-horizon

    Limitations stated: struggles with unseen objects; pose errors accumulate in locomotion, limiting long-horizon navigation; without failure data, generations show optimistic bias toward success.1

Details

About

What it is
In-house test Inferred1
More

Classified by the Atlas from how the authors describe and distribute it.

Built by
1X World Model Team (contributors Daniel Ho, Jack Monas, Juntao Ren, Christina Yu), 1X Technologies1
Released
2025-06 (blog '1X World Model', 2025-06-16; PDF report linked from it)2
Version
Unknown
More

Report carries no version or date.

Last update
Unknown
More

Checked blog page and 1X challenge Space links.

Status
Unknown

Setup

Runs in
AI simulator1
More

Policies are run inside the learned model; e.g. Arcade run continuously for ten minutes with resets.

Robot
Humanoid1
Setting
Unknown
More

Report mentions home use as goal; task settings (air fryer, object drop station, shelf) not tied to a scene type.

Size
Tasks Airfryer, Arcade, Shelf; ~216M Shelf video tokens -> alignment 63.06%; + ~1.46B Arcade tokens -> 71.17%.1

Scoring and access

Scored by
Success rate, Automatic judge1
Leaderboard
None2
Code licence
Unknown
More

No code or weights linked from the blog or report.

Data licence
Unknown
More

Evaluation data not released.

Access
Closed Inferred2

Sources 2

  1. 1https://www.1x.tech/1x-world-model.pdfOfficial blog · checked 10 Oct 2026
  2. 21X World Model | 1XOfficial blog · checked 10 Oct 2026

Change history

  1. Created as a basic entry: identity facts checked at primary sources (phase 1 re-verification).