VLA Gallery / All models / New in 2026

G0Tiny

Galaxea AI (OpenGalaxea) · 2026-02

VLANew in 2026250MNot disclosed

Galaxea's 250M lightweight model, using a SmolVLM2 backbone.

Paper

No paper · Related Galaxea Open-World Dataset and G0 Dual-System VLA Model

Overview

Category

New in 2026

Type

VLA

Released

2026-02

Organization

Galaxea AI (OpenGalaxea)

Open weights

Yes

Parameters

250M

Architecture

Backbone LLM

SmolLM2 · SmolVLM2 config

Vision encoder

—

Action head

Not disclosed (Not disclosed)

Chunk

Not disclosed

Control rate

Not disclosed

Architecture (original summary)

SmolVLM2 backbone; action representation not disclosed

Training

Pretraining accelerators

Not disclosed

Pretraining duration

Not disclosed

Pretraining steps & batch

Not disclosed

GPU/TPU-hours

Not disclosed

Fine-tuning

Not disclosed

Data type

Unverified (notes give scale only)

Data scale

1k hours

Main datasets

Not disclosed

Tasks & evaluation

Not disclosed

Robot

Not disclosed

Inference

Inference device & speed

Not disclosed

Inference memory

Not disclosed

Weights bf16 (calc.)

0.5 GB

Where it runs

Not disclosed

Reinforcement learning & base

Reinforcement learning

Not used · Real-robot data pretraining (supervised)

Base control

Fixed · No chassis control (R1 Pro outputs only EE pose·torso; in demos the chassis is parked aligned to the table)

Notes

Key features

Galaxea's 250M lightweight model, using a SmolVLM2 backbone. Pretrained on 1k hours of data.

Notes

Released 2026-02-12 (README). The note does not describe architecture or benchmarks.

Verification

Confirmed