Hugging Face (LeRobot)
SmolVLA
450M Commercial terms unverified 7/8 verified
1 of 8 required fields are unverified. They are shown as badges, not filled with defaults. Sources for verified fields are listed below.
Facts
- Released
- 2025-06 4 Source
- Parameters
- 0.45 B 3 Source
- Architecture
- VLM + action expert 0 Source
- License
- Weights model card has no licence tag (LeRobot code repository Apache-2.0) 1 Source
- Commercial
- Commercial terms unverified 0 Source
- Weights
- huggingface.co ↗ 3 Source
- Code
- github.com ↗ 1 Source
- Paper
- arxiv.org ↗ 1 Source
- Input modalities
- rgb · language · state 0 Source
- Action space
- Joints (mainly low-cost arms) 0 Source
- Cross-embodiment
- Mainly community embodiments such as SO-100/101 0 Source
- Pretraining data (hours)
- Unverified 0 Source
- Target forms
- Single arm 0 Source
Sources · 6 sources
params_b,release_date,weights_url https://huggingface.co/api/models/lerobot/smolvla_base Model hub page · Fetched 2026-09-03 · No snapshot stored yet safetensors 450,046,176; created 2025-06-01; no license tag
params_b https://arxiv.org/html/2506.01844v1 Paper / official docs · Fetched 2026-09-03 · No snapshot stored yet 450 million parameters
paper_url,release_date https://arxiv.org/abs/2506.01844 Paper / official docs · Fetched 2026-09-03 · No snapshot stored yet Committed 2025-06-02
release_date,weights_url https://huggingface.co/blog/smolvla Model hub page · Fetched 2026-09-03 · No snapshot stored yet June 3, 2025; lerobot/smolvla_base
code_url https://github.com/huggingface/lerobot Official page · Fetched 2026-09-03 · No snapshot stored yet LeRobot code Apache 2.0
params_b,release_date,weights_url,license https://huggingface.co/api/models/lerobot/smolvla_base Model hub page · Fetched 2026-09-04 · No snapshot stored yet safetensors 450,046,176; createdAt 2025-06-01; model card has no license field
Activity signals
Public counters fetched daily from GitHub, Hugging Face and ModelScope. Numbers only; no judgement. · fetched 2026-10-18
GitHub stars
28,271
+663 / 30 d
Commits · 90 d
100+
last push 1 d ago
HF downloads · 30 d
150,301
all-time 831,237 · 472 likes
Weights last modified
2026-10-06
13 d ago
GitHubhuggingface/lerobot · 5,907 forks · 1,031 open issues · latest release v0.6.1 (2026-08-03)
Hugging Facelerobot/smolvla_base
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 |
|---|---|---|---|
| LIBEROSmolVLA 0.24B | avg | 82.8 | paper · Table 2, row 'SmolVLA (0.24B)', LIBERO Avg (87 / 93 / 88 / 63) |
| LIBEROSmolVLA 0.45B (main) | avg | 87.3 | paper · Table 2, row 'SmolVLA (0.45B)', LIBERO Avg |
| LIBEROSmolVLA 2.25B | avg | 88.8 | paper · Table 2, row 'SmolVLA (2.25B)', LIBERO Avg (Spatial 93 / Object 94 / Goal 91 / Long 77) |
| LIBEROSmolVLA 0.45B (main) | goal | 92 | paper · Table 2, row 'SmolVLA (0.45B)', LIBERO Goal |
| LIBEROSmolVLA 0.45B (main) | long | 71 | paper · Table 2, row 'SmolVLA (0.45B)', LIBERO Long |
| LIBEROSmolVLA 0.45B (main) | object | 96 | paper · Table 2, row 'SmolVLA (0.45B)', LIBERO Object |
| LIBEROSmolVLA 0.45B (main) | spatial | 90 | paper · Table 2, row 'SmolVLA (0.45B)', LIBERO Spatial |
| Meta-WorldSmolVLA 0.45B (main) | avg | 57.3 | paper · Table 2, row 'SmolVLA (0.45B)', Meta-World Avg |
| Meta-WorldSmolVLA 0.45B (main) | easy | 82.5 | paper · Table 2, row 'SmolVLA (0.45B)', Meta-World Easy |
| Meta-WorldSmolVLA 0.45B (main) | hard | 45 | paper · Table 2, row 'SmolVLA (0.45B)', Meta-World Hard |
| Meta-WorldSmolVLA 0.45B (main) | medium | 41.8 | paper · Table 2, row 'SmolVLA (0.45B)', Meta-World Medium |
| Meta-WorldSmolVLA 0.45B (main) | very_hard | 60 | paper · Table 2, row 'SmolVLA (0.45B)', Meta-World Very Hard |
| Real-robot evaluation (self-reported)SmolVLA 0.45B (main) | so100_multitask_3_tasks_avg | 78.3 | paper · Table 3 (real-world SO-100), row 'SmolVLA (0.45B)', Avg (Pick-Place 75 / Stacking 90 / Sorting 70) |
| Real-robot evaluation (self-reported)SmolVLA 0.45B (main) | so101_pick_place_lego_in_distribution | 90 | paper · Table 4 (real-world SO-101), row 'SmolVLA (0.45B)', In Distribution |
Deployment recipe
Environment, weights, run command and the pitfalls we hit, exactly as executed on rented GPUs (venv lerobot, 2.0 GB of weights, 2 pitfalls).
Open recipe →Record history
2026-09-03
Sources fetched
params_b,release_date,weights_url · params_b · paper_url,release_date · release_date,weights_url · code_url
2026-09-04
Sources fetched
params_b,release_date,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) | 170 | 197 | 0.91 GB | 0.079 | 1 Preliminary (n<3) Maintainer measured |
NVIDIA RTX 5090 ¥2.4/h · Runyun 2.4 / GongJi 2.5 / Paratera Cloud 2.98 (2026) | 166 | 170 | 0.91 GB | 0.110 | 1 Preliminary (n<3) Maintainer measured |
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 | 226 | 235 | 0.9 GB | — no rental price for this hardware | 1 Preliminary (n<3) Maintainer measured |
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 | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SmolVLA |
Legend Officially supportedCommunity reproducedFine-tunable in theoryUnsupportedUnknown Vendor-declared target form 6 / 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.