As of: 2026-11-13 13:49 CET.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/
Canonical page: https://postcutoff.com/m/xiaomi-robotics-u0/

# Xiaomi-Robotics-U0 (38B) / U0-4B

Xiaomi, Xiaomi-Robotics family. Status: current. Type: world model. Released: 2026-07-13.

## How to call it

| Provider | Model id | Endpoint | Docs |
|---|---|---|---|
| Hugging Face | `XiaomiRobotics/Xiaomi-Robotics-U0` |  | https://huggingface.co/XiaomiRobotics/Xiaomi-Robotics-U0 |
| Hugging Face | `XiaomiRobotics/Xiaomi-Robotics-U0-4B` |  | https://huggingface.co/XiaomiRobotics/Xiaomi-Robotics-U0-4B |
| GitHub |  |  | https://github.com/XiaomiRobotics/Xiaomi-Robotics-U0 |
| ModelScope |  |  | https://modelscope.cn/collections/XiaomiRobotics/Xiaomi-Robotics-U0 |

## Pricing

- Price not published.
- Last checked: 2026-09-29

## Specs

- Input: text, image
- Output: image, video, text
- Context window: not published
- Max output: not published
- Open weights: yes
- Licence: apache-2.0

## Where it stands

- Newer world models in the registry since its release: 1:
  - Genie 4 (2026-10-27) https://postcutoff.com/m/genie-4/
- Weights are public under a permissive licence (Apache-2.0).

## What stands out

- Unified embodied synthesis (first): One autoregressive model (shared discrete visual tokenizer, next-token objective, initialized from Emu3.5) does text-to-image, image editing, multi-view robot scene generation, embodied transfer (editing scenes while keeping multi-view consistency) and embodied video rollout. Source: https://arxiv.org/abs/2607.11643
- Data engine for VLAs: Synthetic data from U0 raised π0.5's out-of-distribution success on hard real-world manipulation tasks from 36.9% to 63.2% (authors). Source: https://arxiv.org/abs/2607.11643
- FlashAR fast decoding: Anti-diagonal grouped visual-token decoding plus vLLM batching: 5.44 s per 1024x1024 image on one H20, 82.86x faster than eager AR. Source: https://huggingface.co/XiaomiRobotics/Xiaomi-Robotics-U0-4B

## Notes

'first' flag is Xiaomi's claim (first model with high-quality multi-view scene generation across multiple robot embodiments). Paper says 38B params; the HF README table says 34B. U0 and U0-FlashAR weights 2026-07-13; U0-4B, U0-Sequence and U0-4B-Sequence weights plus FSDP training code 2026-09-08. U0-Video announced as coming soon. Authors report beating GPT-Image-2.0 in human evals of embodied scene generation/transfer and #1 on World Arena for embodied video. Not an action model: it generates observations/data, not motor commands.

Xiaomi's open embodied world model / synthetic-data engine, the generation-side companion of its VLAs ([xiaomi-robotics-1](xiaomi-robotics-1.md)).

Sources: [arXiv 2607.11643](https://arxiv.org/abs/2607.11643), [HF U0-4B card](https://huggingface.co/XiaomiRobotics/Xiaomi-Robotics-U0-4B), [project page](https://robotics.xiaomi.com/xiaomi-robotics-u0.html).

Model page: https://postcutoff.com/m/xiaomi-robotics-u0/