1X World Model Challenge (1xgpt)

aid: 1x-technologies:1xgpt-world-model-challenge
name: 1X World Model Challenge (1xgpt)
description: 1xgpt is 1X Technologies' open-source world-modeling challenge for humanoid robots, providing
  dataset tooling, baseline code, and evaluation utilities for training and benchmarking generative video
  and world models on humanoid-robot data. It is distributed as a Python research codebase under Apache-2.0
  on GitHub rather than as a hosted API; participants run the code locally to train and evaluate models.
humanURL: https://github.com/1x-technologies/1xgpt
baseURL: https://github.com/1x-technologies/1xgpt
tags:
- World Models
- Humanoid Robots
- Generative Video
- Benchmark
- Open Source
- Research Challenge
- Python
properties:
- type: Repository
  url: https://github.com/1x-technologies/1xgpt
- type: License
  url: https://github.com/1x-technologies/1xgpt/blob/main/LICENSE
- type: README
  url: https://github.com/1x-technologies/1xgpt/blob/main/README.md
- type: HuggingFaceDataset
  url: https://huggingface.co/1x-technologies
features:
- name: Open Dataset
  description: Robot-collected video data released for training world models.
- name: Baseline Models
  description: Reference world-model implementations to benchmark against.
- name: Evaluation Tooling
  description: Scripts and metrics for scoring generated future frames.
- name: Tokenizer
  description: Provided tokenizers for converting frames to discrete tokens.
- name: Reproducible Recipes
  description: Training scripts and configurations for reproducible runs.
- name: Apache 2.0 License
  description: Permissive license enabling research and commercial use.
useCases:
- name: World Model Research
  description: Train generative video models on real humanoid-robot data.
- name: Benchmarking
  description: Compare new model architectures against published baselines.
- name: Robot Learning Curriculum
  description: Use as teaching material for embodied AI courses.
- name: Sim-to-Real Studies
  description: Study learned dynamics against real robot data.
integrations:
- name: PyTorch
- name: Hugging Face Hub
- name: Hugging Face Datasets
- name: NVIDIA CUDA
- name: Weights and Biases
- name: GitHub Actions
authentication:
- type: None
  description: Public open-source repository; no authentication required.