Physical Intelligence
π0 (pi-zero)
3.3B 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
- 2025-02 1 Source
- Parameters
- 3.3 B 1 Source
- Architecture
- VLM + action expert 0 Source
- License
- Apache-2.0 (openpi code); no separate licence statement for weights, repository includes Gemma Terms of Use 3 Source
- Commercial
- Commercial terms unverified 0 Source
- Weights
- huggingface.co ↗ 2 Source
- Code
- github.com ↗ 1 Source
- Paper
- arxiv.org ↗ 1 Source
- Input modalities
- rgb · language · state 0 Source
- Action space
- Joints (mainly arms/dual-arm) 0 Source
- Cross-embodiment
- Multi-embodiment pretraining Unverified 0 Source List unverified
- Pretraining data (hours)
- Unverified 0 Source
- Target forms
- Single arm · Dual arm · Mobile manipulator 0 Source
Sources · 6 sources
license https://raw.githubusercontent.com/Physical-Intelligence/openpi/main/LICENSE Official page · Fetched 2026-09-03 · No snapshot stored yet Code Apache-2.0
license https://raw.githubusercontent.com/Physical-Intelligence/openpi/main/LICENSE_GEMMA.txt Official page · Fetched 2026-09-03 · No snapshot stored yet Gemma Terms of Use govern model derivatives
weights_url,code_url https://raw.githubusercontent.com/Physical-Intelligence/openpi/main/README.md Official page · Fetched 2026-09-03 · No snapshot stored yet Official weights gs://openpi-assets/checkpoints/pi0_base; README has no weights-licence section
release_date https://www.pi.website/blog/openpi Official page · Fetched 2026-09-03 · No snapshot stored yet Open Sourcing π0, February 4, 2025
params_b,paper_url https://arxiv.org/html/2410.24164v1 Paper / official docs · Fetched 2026-09-03 · No snapshot stored yet total 3.3 billion parameters
weights_url,license https://huggingface.co/api/models/lerobot/pi0_base Model hub page · Fetched 2026-09-03 · No snapshot stored yet LeRobot port license tagged gemma
Activity signals
Public counters fetched daily from GitHub, Hugging Face and ModelScope. Numbers only; no judgement. · fetched 2026-10-18
GitHub stars
14,290
+413 / 30 d
Commits · 90 d
1
last push 56 d ago
HF downloads · 30 d
17,679
all-time 134,099 · 29 likes
Weights last modified
2026-10-06
13 d ago
GitHubPhysical-Intelligence/openpi · 2,568 forks · 363 open issues · no GitHub release
Hugging Facelerobot/pi0_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 |
|---|---|---|---|
| LIBEROπ0 (fine-tuned per suite) | avg | 94.2 | third-party · OpenVLA-OFT paper, Table I, row 'π0 (fine-tuned)', Average |
| LIBEROπ0-FAST (π0 + FAST tokenizer, fine-tuned) | avg | 85.5 | third-party · OpenVLA-OFT paper, Table I, row 'π0 + FAST (fine-tuned)', Average (Spatial 96.4 / Object 96.8 / Goal 88.6 / Long 60.2) |
| LIBEROπ0 (fine-tuned per suite) | goal | 95.8 | third-party · OpenVLA-OFT paper, Table I, row 'π0 (fine-tuned)', Goal |
| LIBEROπ0 (fine-tuned per suite) | long | 85.2 | third-party · OpenVLA-OFT paper, Table I, row 'π0 (fine-tuned)', Long |
| LIBEROπ0 (fine-tuned per suite) | object | 98.8 | third-party · OpenVLA-OFT paper, Table I, row 'π0 (fine-tuned)', Object |
| LIBEROπ0 (fine-tuned per suite) | spatial | 96.8 | third-party · OpenVLA-OFT paper, Table I, row 'π0 (fine-tuned)' (multi-input setup, modified dataset), Spatial |
| Real-robot evaluation (self-reported) | agibot_g1_5_tasks_avg_completion_score | 0.6 score 0-1 | third-party · GO-1 paper, Fig. 5(b) 'Average' bar label for π0 |
| Real-robot evaluation (self-reported)π0 (3.5B), multi-task fine-tuned | so100_multitask_3_tasks_avg | 61.7 | third-party · SmolVLA paper, Table 3, row 'π0 (3.5B)', Avg (Pick-Place 100 / Stacking 40 / Sorting 45) |
| RoboTwin 2.0 | easy | 46.4 | third-party · X-VLA paper, Table 2, row 'π0', RoboTwin-2.0 Easy (Hard 16.4) |
| SimplerEnv · Google Robot | variant_aggregation | 56.8 | third-party · X-VLA paper, Table 2, row 'π0', Simpler VA |
| SimplerEnv · Google Robot | visual_matching | 58.8 | third-party · X-VLA paper, Table 2, row 'π0 (Black et al., 2024a)', Simpler VM |
| SimplerEnv · WidowX (Bridge) | visual_matching | 27.8 | third-party · X-VLA paper, Table 2, row 'π0', Simpler WidowX |
| VLABench | avg_progress_score | 37.8 Avg. PS | third-party · X-VLA paper, Table 2, row 'π0', VLABench Avg. PS |
Deployment recipe
Not reproduced yet · Needs the Google PaliGemma tokenizer, which is gated on Hugging Face (licence acceptance required); not reachable through the mirror route.
Record history
2026-09-03
Sources fetched
license · license · weights_url,code_url · release_date · params_b,paper_url · 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) | — | — | — | — no latency data | No data |
NVIDIA RTX 5090 ¥2.4/h · Runyun 2.4 / GongJi 2.5 / Paratera Cloud 2.98 (2026) | — | — | — | — no latency data | No data |
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 | — | — | — | — no latency data | No data |
No latency measurements for this model yet · Data comes from literature, maintainer runs, or crowd CLI uploads. None exist so far.
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 | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| π0 (pi-zero) |
Legend Officially supportedCommunity reproducedFine-tunable in theoryUnsupportedUnknown Vendor-declared target form 3 / 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.