OpenDriveLab (HKU et al.)
UniVLA
7.54B Commercial use OK 8/8 verified
Facts
- Released
- 2025-05 1 Source
- Parameters
- 7.54 B 1 Source
- Architecture
- VLM + action expert 0 Source
- License
- Apache-2.0 (tagged on both code repository and model card) 2 Source
- Commercial
- Commercial use OK 0 Source
- Weights
- huggingface.co ↗ 1 Source
- Code
- github.com ↗ 1 Source
- Paper
- arxiv.org ↗ 1 Source
- Input modalities
- rgb · language 0 Source
- Action space
- Latent actions → decoded per embodiment 0 Source
- Cross-embodiment
- Open X-Embodiment + Ego4D human-video subset; LIBERO / CALVIN / SimplerEnv / navigation / AgileX real robot 1 Source
- Pretraining data (hours)
- Unverified 0 Source
- Target forms
- Single arm · Dual arm · Wheeled 0 Source
Architecture · Prismatic 7B backbone + task-centric latent actions (two-stage VQ-VAE) + approx. 12M-parameter action decoder
Sources · 3 sources
license,release_date,arch_notes,cross_embodiment,code_url https://github.com/OpenDriveLab/UniVLA Official page · Fetched 2026-09-04 · No snapshot stored yet README: Apache-2.0; code released 2025-05; 7B backbone, action decoder around 12M; OXE + Ego4D; AgiLex real robot
params_b,weights_url,license https://huggingface.co/api/models/qwbu/univla-7b Model hub page · Fetched 2026-09-04 · No snapshot stored yet safetensors 7,541,237,184；license: apache-2.0；createdAt 2025-05-06
paper_url https://arxiv.org/abs/2505.06111 Paper / official docs · Fetched 2026-09-04 · No snapshot stored yet UniVLA: Learning to Act Anywhere with Task-centric Latent Actions
Activity signals
Public counters fetched daily from GitHub, Hugging Face and ModelScope. Numbers only; no judgement. · fetched 2026-10-31
GitHub stars
1,142
+7 / 30 d
Commits · 90 d
0
last push 347 d ago
HF downloads · 30 d
81
all-time 9,952 · 10 likes
Weights last modified
2025-08-02
456 d ago
ModelScopeWangC23/univla-7b · 41 downloads · updated 2025-08-11 · same-name mirror, not the vendor org
Hugging Faceqwbu/univla-7b
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 |
|---|---|---|---|
| CALVIN | abc_d | 3.8 avg_len | paper · Table A-II, row 'UniVLA (Ours)', Avg. Len. |
| LIBEROUniVLA (Bridge) | avg | 92.5 | paper · Table I, row 'UniVLA (Ours)' / 'Bridge', Average |
| LIBEROUniVLA (Full) | avg | 95.2 | paper · Table I, row 'UniVLA (Ours)' / 'Full', column Average |
| LIBEROUniVLA (Human) | avg | 88.7 | paper · Table I, row 'UniVLA (Ours)' / 'Human', Average |
| LIBEROUniVLA (Full) | goal | 95.6 | paper · Table I, row 'UniVLA (Ours)' / 'Full', column Goal |
| LIBEROUniVLA (Full) | long | 92 | paper · Table I, row 'UniVLA (Ours)' / 'Full', column Long |
| LIBEROUniVLA (Full) | object | 96.8 | paper · Table I, row 'UniVLA (Ours)' / 'Full', column Object |
| LIBEROUniVLA (Full) | spatial | 96.5 | paper · Table I, row 'UniVLA (Ours)' / 'Full', column Spatial |
| Real-robot evaluation (self-reported)UniVLA (Full) | avg | 81.7 | paper · Table A-V, row 'UniVLA (Full)', Average Succ. |
| Real-robot evaluation (self-reported) | generalizability_avg | 68.9 | paper · Table II, row 'UniVLA (Ours)', Average Succ. |
| SimplerEnv · WidowX (Bridge) | overall | 47.9 | paper · Table A-III, row 'UniVLA', #Overall |
| SimplerEnv · WidowX (Bridge)UniVLA (Decoder-only) | overall | 35.4 | paper · Table A-III, row 'UniVLA (Decoder-only)', #Overall |
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
2026-09-04
Sources fetched
license,release_date,arch_notes,cross_embodiment,code_url · params_b,weights_url,license · paper_url
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 | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| UniVLA |
Legend Officially supportedCommunity reproducedFine-tunable in theoryUnsupportedUnknown Vendor-declared target form 2 / 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.