ModelFeatured
Octo
Generalist robot policy Transformer with multi-modal inputs and multi-embodiment finetuning.
93MFully openData: Open X-EmbodimentUpdated 2024-07-31
ModelFeaturedWeights open
π0 / pi0 (Physical Intelligence)
Physical Intelligence generalist VLA (~3B) with flow-matching actions; π0 weights are partially open — strong dexterous-manipulation reference in the π0 family.
Why we recommend it
pi0 is a strong open generalist VLA checkpoint often discussed with openpi tooling. Choose it when you need a modern baseline and can host the stack. Heavier than SmolVLA; check release notes and licenses. Intermediate-to-advanced.
Overview
π0 (pi0) from Physical Intelligence is a generalist robot foundation model using flow-matching style action generation. It represents the “frontier generalist policy” wave: broad task coverage, multi-embodiment pretaining, and strong real-world demos.
Openness is typically weights-oriented rather than fully reproducible industrial training stacks. Always read the model card for data, license, and intended use. Pair with openpi for runnable training/inference code paths.
Who it is for
Teams evaluating frontier generalist policies; advanced VLA practitioners.
Key highlights
- Generalist real-robot focus
- Flow-matching action modeling
- Paired open code via openpi
When to use
- Finetuning a strong generalist on your embodiment
- Systems research on foundation policies
When not to use
- First-week beginners without GPU budget
Getting started
- 1Check license and weight access
- 2Run openpi inference
- 3Small supervised finetune on curated demos
Papers & reading
How it compares
Frontier VLA trajectory; use OpenPI for open inference code paths.
Caveats & pitfalls
- License may restrict commercial products.
- Data mixture opacity can hurt scientific claims.
Content reviewed 2026-07-23