Spirit AIChina2026

Spirit-v1.5

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

3 of 6 stated · 3 gaps

Model type ⁠A vision-language-action
Controls not stated
Inputs not stated
Access ⁠C open download
Parameters not stated
Base model ⁠B Qwen3-VL-4B-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: "Spirit-v1.5 is a Vision-Language-Action (VLA) model designed specifically for robotic control.".

Controls

Note The README describes the output as an action chunk: "The model accepts current observations and textual descriptions as input and generates the next action chunk for the robot to execute."

Inputs

Note README: "The model accepts current observations and textual descriptions as input and generates the next action chunk for the robot to execute."

Access

Spirit AI: "we have open-sourced the Spirit-v1.5 model weights and codebase".

Parameters

Note The model card names the backbone: "A VLM backbone: `Qwen/Qwen3-VL-4B-Instruct`".

Base model

Model card: "A VLM backbone: `Qwen/Qwen3-VL-4B-Instruct`", with a diffusion-transformer action head.

Full specifications

6 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`	A Developer wording: "Spirit-v1.5 is a Vision-Language-Action (VLA) model designed specifically for robotic control.".
Inputs	not stated	Note README: "The model accepts current observations and textual descriptions as input and generates the next action chunk for the robot to execute." ⁠A
Controls	not stated	Note The README describes the output as an action chunk: "The model accepts current observations and textual descriptions as input and generates the next action chunk for the robot to execute." ⁠A
Degrees of freedom controlled	not stated
Size and speed
Parameters	not stated	Note The model card names the backbone: "A VLM backbone: `Qwen/Qwen3-VL-4B-Instruct`". ⁠B
Control rate	not stated	Note The README gives an action chunk length, not a rate: "Action chunk size (Default: 60)." ⁠A
Where it runs and what it is built on
Runs where	not stated	Note README: "Tested on NVIDIA A100 80GB GPU." ⁠A
Base model	Qwen3-VL-4B-Instruct	B Model card: "A VLM backbone: `Qwen/Qwen3-VL-4B-Instruct`", with a diffusion-transformer action head.
Openness
Access	open download	C Spirit AI: "we have open-sourced the Spirit-v1.5 model weights and codebase".
License	Apache-2.0	B The model card header reads "license: apache-2.0". The code repository is under a different licence: "MIT License".
The developer’s own claims
Learns new tasks	needs retraining	C Developer claim: "The model learns a universal policy that can be adapted to new tasks in much fewer steps." In its benchmark "all models are finetuned on the same set of data for Table30 tasks".
Works across robots	yes	C Developer claim: "We evaluate it on a recently established real-world robot benchmark, RoboChallenge Table30". "These tasks are evaluated across multiple embodiments, including single-arm manipulators (Franka, Arx5 and UR5) and a dual-arm system (ALOHA)". 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

Spirit AI describes Spirit-v1.5 as a vision-language-action model designed specifically for robotic control.
Spirit AI released the base model checkpoint, a fine-tuned checkpoint and the inference code in January 2026.

Sources

A Manufacturer product page. github.com/Spirit-AI-Team/spirit-v1.5 · Retrieved 2026-10-14
B huggingface.co/Spirit-AI-robotics/Spirit-v1.5 · Retrieved 2026-10-14
C www.spirit-ai.com/en/blog/spirit-v1-5 · 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.