← Leaderboard
π0-FAST
VLAPhysical Intelligence · Jan 16, 2025
Autoregressive variant of π0 that predicts actions with the FAST frequency-space action tokenizer. Its DROID checkpoint is ranked on RoboArena.
- Type
- VLA
- Parameters
- 3B
- Backbone
- PaliGemma 3B
- Weights
- Open
- License
- Gemma
- Commercial use
- Unclear
- Released
- Jan 16, 2025
- Robots
- DROID Franka
- SO-101
- –
- LeRobot
- –
Benchmark results
4 of 13 benchmarksRoboLab-120Simulation
- Official leaderboard ↗Nov 17, 2026· n = 1200
#18 of 2215.5%
LIBERO-PlusSimulation
- Paper / model card ↗Oct 2025
#22 of 2462.5%, self-reported by the authors, not a leaderboard entry
- Paper / model card ↗Feb 4, 2025
#44 of 4685.5%, self-reported by the authors, not a leaderboard entry
RoboArenaReal robot
- Official leaderboard ↗Nov 17, 2026· n = 941
#5 of 131582
Not reported on 9 benchmarks
Robustness drop: 85.5% on LIBERO → 62.5% on LIBERO-Plus. The perturbed variants show how much of the LIBERO score is memorization.
Adoption
updated Nov 17, 2026Running the model
| Hardware | Precision | Config | Latency / chunk | Steps | VRAM | Chunk | Control | Source |
|---|---|---|---|---|---|---|---|---|
| L40S | – | RoboLab-120 evaluation (DROID) | – | – | – | 10 | – | May 9, 2026 ↗ |
| RTX 4090 | – | autoregressive FAST decoding | 750 ms | – | – | – | – | Jan 16, 2025 ↗ |
Runs on
Wrong or missing number? Report an error