SmolVLA
450M-parameter vision-language-action model for affordable robotics
About
SmolVLA is Hugging Face's 450M-parameter vision-language-action model for affordable robotics, released June 2025 with weights, code, and training data all open. It pairs a small vision-language backbone with a flow-matching action expert and was trained entirely on community LeRobot datasets, reaching 87.3% on LIBERO per the team's report — competitive with models ten times its size. It's aimed at low-cost arms like the SO-100 series; dexterous or bimanual tasks are beyond what it was built for.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Team-reported: 87.3% on LIBERO (vs OpenVLA 7B at 76.5%); community pretraining lifted SO-100 success from 51.7% to 78.3%.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified tier, metrics-v1). Evidence: Released 2025-06-03 by Hugging Face (LeRobot team); 450M params (SmolVLM2 + flow-matching action expert); Trained entirely on community LeRobot data (10.6M frames); paper arXiv:2506.01844; LIBERO 87.3% and SO-100 figures are team-reported and labeled as such; Exact HF card license text not re-read this session; LeRobot repo is Apache-2.0 — confirm on the card before production use. License: Released fully open (weights, code, data); LeRobot repo is Apache-2.0 — confirm exact card license before production use..
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 8.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 6.0/10
- · Released 2025-06-03 by Hugging Face (LeRobot team); 450M params (SmolVLM2 + flow-matching action expert)
- · Trained entirely on community LeRobot data (10.6M frames); paper arXiv:2506.01844
- · LIBERO 87.3% and SO-100 figures are team-reported and labeled as such
- · Exact HF card license text not re-read this session; LeRobot repo is Apache-2.0 — confirm on the card before production use