Skip to content
embodied.guide

Back to the AI models

Tsinghua AIRChina2025

X-VLA

An AI model: a vision-language-action model (VLA) takes camera images and an instruction and turns them into movements for a robot. This page describes the software, not a robot.

RobotsG1by Agibot C

Key figures as stated by the developer

6 of 6 stated
  • Model type ⁠D vision-language-action
  • Controls ⁠A arms and hands
  • Inputs ⁠A camera language robot state
  • Access ⁠C open download
  • Parameters ⁠B approximately 900,000,000
  • Base model ⁠B Florence-2-large
How to read the figures
14kgA30.9lbA
The letter in the grey square after a figure points to the source the figure came from. Robots with several sources carry several letters.
74.5lb
Imperial figures are our own conversion unless the manufacturer states an imperial figure itself. The source letter always belongs to the metric figure — a conversion has no source of its own.

The four states

Versions and notes in the key figures (6)

Model type

Developer wording: "X-VLA: Soft-Prompted Transformer as Scalable Cross-Embodiment Vision-Language-Action Model".

Controls

Paper: "we standardize the action space into end-effector (EEF) pose representation".

Inputs

Paper: "simultaneously encoding multi-view images, language prompts, and proprioceptive features".

Access

Repository: "The X-VLA-Pt model is the foundation checkpoint, trained across multiple robot domains."

Parameters

Model card title: "X-VLA 0.9B (Foundation Edition)"; its table gives the parameters as ≈ 0.9 B.

Base model

Model card header: "base_model: - microsoft/Florence-2-large".

Full specifications

9 of 12 fields stated
The letter points to the source at the foot of the page. Fields without a figure are dashed: the developer does not state them — that is not the same as zero, and not the same as no.
Field Value Source and caveats
Input and output
Model typevision-language-actionD

Developer wording: "X-VLA: Soft-Prompted Transformer as Scalable Cross-Embodiment Vision-Language-Action Model".

Inputscamera language robot stateA

Paper: "simultaneously encoding multi-view images, language prompts, and proprioceptive features".

Controlsarms and handsA

Paper: "we standardize the action space into end-effector (EEF) pose representation".

Degrees of freedom controllednot stated
Size and speed
Parametersapproximately 900,000,000B

Model card title: "X-VLA 0.9B (Foundation Edition)"; its table gives the parameters as ≈ 0.9 B.

Control ratenot stated
Where it runs and what it is built on
Runs wherenot stated
Base modelFlorence-2-largeB

Model card header: "base_model: - microsoft/Florence-2-large".
Openness
Accessopen downloadC

Repository: "The X-VLA-Pt model is the foundation checkpoint, trained across multiple robot domains."

LicenseApache 2.0B

Model card header: "license: apache-2.0". The repository README says "This repository is licensed under the Apache License 2.0."
The developer’s own claims
Learns new tasksneeds retrainingA

Developer claim: "the policy is then effectively specialized to the new embodiment through finetuning."

Works across robotsyesA

Developer claim: "the model learns an embodiment-agnostic generalist policy." The pre-training data are "spanning seven platforms across five types of robotic arms, ranging from single-arm to bi-manual setups."

Yes only if the developer itself says the model runs on different robot models, for example robots from two manufacturers or two of one manufacturer’s models. Several units of the same robot do not count.

Notes

  • X-VLA is a flow-matching vision-language-action model from the Institute for AI Industry Research (AIR), Tsinghua University, and Shanghai Artificial Intelligence Laboratory.
  • The Institute for AI Industry Research is in Tsinghua Science and Technology Park, Haidian District, Beijing.
  • The repository states that X-VLA has been accepted to ICLR 2026.

Sources

One line per source, not per figure: a source has one retrieval date. The letter beside a figure points to the line it came from.