Galaxea
Galaxea G0.5
2B Non-commercial 8/8 verified
Facts
- Released
- 2026-06 1 Source
- Parameters
- 2 B 0 Source
- Architecture
- Monolithic autoregressive 0 Source
- License
- G0.5 Community License (non-commercial + limited patent licence; model card license: other); versions before 2026-01-04 were Apache-2.0 2 Source
- Commercial
- Non-commercial 0 Source
- Weights
- huggingface.co ↗ 1 Source
- Code
- github.com ↗ 1 Source
- Paper
- arxiv.org ↗ 0 Source
- Input modalities
- rgb · language · state 0 Source
- Action space
- Mobile manipulator (chassis + dual-arm) 0 Source
- Cross-embodiment
- 14 embodiments; R1 Lite / R1 Pro / SO-100/101 / DROID-Franka; Galaxea Open-World Dataset 500+ hours 1 Source
- Pretraining data (hours)
- 500 h 1 Source
- Target forms
- Dual arm, wheeled · Single arm 0 Source
Architecture · Qwen3.5 2B VLM, single autoregressive stream outputs reasoning and actions together; predecessors G0Plus 3B / G0Tiny 250M (2026-01)
Sources · 2 sources
release_date,license,arch_notes,cross_embodiment,pretrain_hours,code_url https://github.com/OpenGalaxea/GalaxeaVLA Official page · Fetched 2026-09-05 · No snapshot stored yet README: G0.5 checkpoint 2026-06-16, paper 2026-08-12; G0.5 Community License (Non-Commercial + Limited Patent License); Qwen3.5 2B; 14 embodiments; 500+ hours
weights_url,license https://huggingface.co/api/models/OpenGalaxea/G05 Model hub page · Fetched 2026-09-05 · No snapshot stored yet Model card license: other (G05-Community-License); createdAt 2026-06-16
Activity signals
Public counters fetched daily from GitHub, Hugging Face and ModelScope. Numbers only; no judgement. · fetched 2026-10-31
GitHub stars
820
+17 / 30 d
Commits · 90 d
1
last push 80 d ago
HF downloads · 30 d
0
all-time 0 · 22 likes
Weights last modified
2026-06-30
124 d ago · gated (request access)
GitHubOpenGalaxea/GalaxeaVLA · 71 forks · 28 open issues · no GitHub release
ModelScopeno same-name mirror found (searched by repo name)
Hugging FaceOpenGalaxea/G05
stars · last 30 d
Public benchmark scores
As printed in the source; protocols differ between papers. Full table with notes on the benchmarks page.
| Benchmark | Subset | Score | Source |
|---|---|---|---|
| LIBERO | avg | 98.9 | paper · Table 3, row 'G0.5 (Ours)', column 'Average' |
| LIBERO | goal | 98.6 | paper · Table 3, row 'G0.5 (Ours)', column 'Goal' |
| LIBERO | long | 98.6 | paper · Table 3, row 'G0.5 (Ours)', column 'Long' |
| LIBERO | object | 100 | paper · Table 3, row 'G0.5 (Ours)', column 'Object' |
| LIBERO | spatial | 98.4 | paper · Table 3, row 'G0.5 (Ours)', column 'Spatial' |
| Other benchmarks | task_success_score | 31.4 | paper · Abstract text: '2025 BEHAVIOR Challenge on 50 long-horizon household mobile manipulation tasks using a generalist policy (31.4% vs. 26.3% for pi0.5 and 26.1% for the challenge winner)'; detailed in Table 4/6 |
| Real-robot evaluation (self-reported) | droid_zero_shot_unseen_env_objects_avg | 82.5 | paper · Abstract / Sec. 5.1.2 text: 'DROID post-training followed by zero-shot transfer to an unseen environment and objects (82.5%)' |
| Real-robot evaluation (self-reported) | r1_lite_r1_pro_finetune_avg | 76.7 | paper · Abstract / Sec. 5.4.2 text: 'real-world fine-tuning on R1-Lite/R1-Pro robots (76.7% vs. 53.3% for pi0.5 and 24.4% for GR00T-N1.7)' |
| RoboTwin 2.0 | avg | 93.3 | paper · Table 2, row 'G0.5 (Ours)', column 'Average' |
| RoboTwin 2.0 | clean | 93.7 | paper · Table 2, row 'G0.5 (Ours)', column 'Clean' |
| RoboTwin 2.0 | randomized | 92.8 | paper · Table 2, row 'G0.5 (Ours)', column 'Rand.' |
| SimplerEnv · WidowX (Bridge) | avg | 87.3 | paper · Table 1 'Results on Bridge-SimplerEnv. All numbers are success rates (%)', row 'G0.5 (Ours)', column Average |
Deployment recipe
No recipe yet. We publish only steps that actually ran on a rented GPU with a measured latency; this model has not been through that yet.
Record history
2026-09-05
Sources fetched
release_date,license,arch_notes,cross_embodiment,pretrain_hours,code_url · weights_url,license
Latency & cost
cost = p50 ms ÷ 1000 × rental ÷ 3600 × 1000
| Hardware | p50 ms | p95 ms | Peak VRAM | Cost per 1k inferences ¥ | samples |
|---|---|---|---|---|---|
NVIDIA RTX 4090 ¥1.68/h · GongJi Compute (2026) | — | — | — | — no latency data | No data |
NVIDIA RTX 5090 ¥2.4/h · Runyun 2.4 / GongJi 2.5 / Paratera Cloud 2.98 (2026) | — | — | — | — no latency data | No data |
NVIDIA A100 80GB no rental price | — | — | — | — no latency data | No data |
NVIDIA Jetson AGX Thor no rental price | — | — | — | — no latency data | No data |
NVIDIA H100 80GB no rental price | — | — | — | — no latency data | No data |
NVIDIA A800 80GB no rental price | — | — | — | — no latency data | No data |
No latency measurements for this model yet · Data comes from literature, maintainer runs, or crowd CLI uploads. None exist so far.
Compatibility
Colored only where we hold evidence. An amber dot means the vendor declares this form as a target; that is a claim, not a verified compatibility.
| Unitree G1 | Unitree H1 | AgiBot Genie G1 | AgileX Piper arm | SO-101 | Franka FR3 / Panda | Fourier GR-2 | WidowX 250 S | Google Robot（Everyday Robots） | ALOHA 2 / Cobot Magic (dual-arm) | Galaxea R1 Pro | Franka Emika Panda | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Galaxea G0.5 |
Legend Officially supportedCommunity reproducedFine-tunable in theoryUnsupportedUnknown Vendor-declared target form 5 / 12 cells with a definite status
Reproduction records
Did this model run on your robot? Report it with a link (log, video or repo). Records are reviewed by the maintainers; a verified “works” or “works after fine-tuning” upgrades the matrix cell to “community verified” with your link as evidence. Anonymous is fine; nothing but the fields below is stored.
No verified record yet for this model.
Clear direction, grounded judgement.