X Square RobotChina2026

Wall-OSS-0.5

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.

Robotsnot stated

Key figures as stated by the developer

5 of 6 stated · 1 gap

Model type ⁠C vision-language-action
Controls not stated
Inputs ⁠A camera language robot state
Access ⁠A open download
Parameters ⁠B 4,000,000,000
Base model ⁠A Qwen2.5-VL-3B-Instruct

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: "an open-source 4B Vision-Language-Action (VLA) foundation model built upon a 3B VLM backbone".

Controls

Note The paper maps the actions of every platform to one shared action format, which is wider than the closed list: "bimanual end-effector poses, joint positions, gripper states, mobile-base motion, lift/waist actuation, and head actuation".

Inputs

Paper: "Four token streams, vision, language, proprioception, and discrete actions, are routed through the VL Expert".

Access

X Square Robot: "We release an open-source 4B VLA"; the GitHub page lists the weights: "WALL-OSS-0.5: https://huggingface.co/x-square-robot/wall-oss-0.5".

Parameters

Abstract: "an open-source 4B VLA built upon a 3B VLM backbone".

Base model

Paper: "Wall-OSS-0.5 is initialized from Qwen2.5-VL-3B-Instruct".

Full specifications

8 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 type	`vision-language-action`	C Developer wording: "an open-source 4B Vision-Language-Action (VLA) foundation model built upon a 3B VLM backbone".
Inputs	`camera` `language` `robot state`	A Paper: "Four token streams, vision, language, proprioception, and discrete actions, are routed through the VL Expert".
Controls	not stated	Note The paper maps the actions of every platform to one shared action format, which is wider than the closed list: "bimanual end-effector poses, joint positions, gripper states, mobile-base motion, lift/waist actuation, and head actuation". ⁠A
Degrees of freedom controlled	not stated	Note The size of the shared action vector, not a figure for a named robot: "a fixed 26-dimensional action representation". ⁠A
Size and speed
Parameters	4,000,000,000	B Abstract: "an open-source 4B VLA built upon a 3B VLM backbone".
Control rate	approximately 21Hz	A Model inference layer, not motor control. "We evaluate the optimized inference stack on a single RTX 5090 with three-view image input"; the stack "reaches approximately 21 Hz and 15 Hz at the two resolutions respectively". The paper also says it "sustains real-time control (15 Hz) at high input resolution".
Where it runs and what it is built on
Runs where	not stated
Base model	Qwen2.5-VL-3B-Instruct	A Paper: "Wall-OSS-0.5 is initialized from Qwen2.5-VL-3B-Instruct".
Openness
Access	open download	A X Square Robot: "We release an open-source 4B VLA"; the GitHub page lists the weights: "WALL-OSS-0.5: https://huggingface.co/x-square-robot/wall-oss-0.5".
License	not stated	Note The code repository is under the "Apache License" (Apache 2.0). ⁠C
The developer’s own claims
Learns new tasks	needs retraining	B Developer claim: "After fine-tuning, the same checkpoint serves as a stronger adaptation prior". Before fine-tuning, "the pretrained checkpoint achieves non-trivial zero-shot real-robot behavior"; the paper adds: "Zero-shot capability has clear limits."
Works across robots	yes	C Developer claim: "Wall-OSS-0.5 supports a wide variety of robot embodiments ranging from commercial platforms to open-source benchmarks". "The model is pretrained across more than 20 embodiments". 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 Square Robot describes Wall-OSS-0.5 as an open-source vision-language-action foundation model built upon a 3B vision-language backbone.
The paper on Wall-OSS-0.5 was submitted to arXiv on 29 May 2026.

Sources

A Manufacturer product page. arxiv.org/html/2605.30877v2 · Retrieved 2026-10-14
B arxiv.org/abs/2605.30877 · Retrieved 2026-10-14
C huggingface.co/x-square-robot/wall-oss-0.5 · Retrieved 2026-10-14

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.