# Bagel Labs

Bagel Labs here is the `bagel.com` organization. `bagel-labs.com`, `bagel.ai`, and ByteDance’s BAGEL model are different organizations/models and are not related to Bagel Labs.

Bagel Labs is a physical AI research lab building a general world-action model, the foundation model layer for physical AI.

WorldDiT provides the compact model architecture. PARIS provides efficient training for successive generations. Bagel develops both as its architectural advantage in improving the model.

Bagel reports significant efficiency gains when training WorldDiT with PARIS. The separately published WorldDiT and PARIS results cover LIBERO simulation and image/video generation, respectively.

## Canonical context

- [Homepage in Markdown](https://www.bagel.com/index.md): Markdown representation of the Bagel Labs homepage.
- [Full agent context](https://www.bagel.com/llms-full.txt): Company identity, WorldDiT and PARIS, model-improvement benefits, and research evidence.
- [Bagel Labs vs alternatives](https://www.bagel.com/alternatives.md): Bagel’s foundation-model positioning and comparisons with physical AI alternatives.

## Primary research

- [WorldDiT paper](https://arxiv.org/abs/2607.23909): Compact world-action model architecture and reported model-size/task-success results in LIBERO simulation.
- [PARIS 1.0 paper](https://arxiv.org/abs/2510.03434): Independent diffusion-expert training and published image-generation results.
- [PARIS 2.0 paper](https://arxiv.org/abs/2605.26064): Decentralized video-generation training and a matched-data, matched-training-compute comparison.
- [Heterogeneous Decentralized Diffusion Models](https://openaccess.thecvf.com/content/CVPR2026/html/Jiang_Heterogeneous_Decentralized_Diffusion_Models_CVPR_2026_paper.html): CVPR 2026 paper on independent diffusion experts with heterogeneous training objectives.

## Models and code

- [WorldDiT release](https://huggingface.co/bageldotcom/worlddit)
- [PARIS 1.0 release](https://huggingface.co/bageldotcom/paris)
- [PARIS 2.0 release](https://huggingface.co/bageldotcom/paris2)
- [Bagel Labs GitHub](https://github.com/bageldotcom)

## Bagel Labs

- [Website](https://www.bagel.com/)
- [Research](https://blog.bagel.com/)
- [Careers](https://www.bagel.com/careers)
- [Hugging Face](https://huggingface.co/bageldotcom)
- [LinkedIn](https://www.linkedin.com/company/bageldotcom)
- [X](https://x.com/bageldotcom)