Robot AI model
SmolVLA
SmolVLA combines a compact vision-language backbone with a flow-matching action expert. Hugging Face trains it on community-shared LeRobot data and reports simulated and real-arm manipulation experiments on SO100 and SO101.
Records last verified Oct 13, 2026 · Send a correction
Also known as SmolVLA-450M
- Organization
- Hugging Face
- Type
- VLA
- Released / introduced
- Jun 2, 2025
- Public weights
- Public weights
- Licence / scope
- Apache-2.0
- Parameters
- 450M
- Base model
- SmolVLM2-500M-Video-Instruct (partial backbone)
- Humanoid association
- None reported
- Inputs
- vision, language, proprioception
- Outputs
- continuous action chunks
Release dates can mark an announcement or paper introduction. Public checkpoints, code licences and weight terms are distinct; follow the publisher’s links for the stated scope.
Documented embodiments
Source-associated hardware; this list does not establish plug-and-play compatibility or commercial robot capability.
- SO100
- SO101
Disclosed training data
- LeRobot community SO100 manipulation datasets
More robot AI models
Evidence
Sources
- [s1] SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics ↗
Hugging Face · paper · published Jun 2, 2025 · accessed Oct 13, 2026
- [s2] SmolVLA: architecture, community data and experiments ↗
Hugging Face · official · published Jun 3, 2025 · accessed Oct 13, 2026
- [s3] SmolVLA base model card ↗
Hugging Face · official · accessed Oct 13, 2026