{"schemaVersion":"0.4.0","type":"product","id":"prd_openpi","slug":"openpi","name":"Physical Intelligence openpi","companyId":"co_physical_intelligence","summary":"Open robot foundation models (π0, π0-FAST and π0.5) that take camera images and an instruction and output robot actions, with code to fine-tune them on a robot maker's own data.","primaryWorld":"ai-infrastructure","worlds":["ai-infrastructure"],"audience":["developer"],"categorySlugs":["ai-infrastructure"],"capabilitySlugs":["manipulation"],"evidenceStage":{"value":"deployment","asOf":"2026-10-18","sourceIds":["src_pi_openpi_blog","src_openpi_repo"],"note":"Model weights and code released for anyone to download.","basis":{"kind":"public_availability","detail":"Physical Intelligence released the π0 and π0-FAST model weights and code on February 4, 2025, and π0.5 in September 2025; anyone can download them through the openpi repository.","sourceIds":["src_pi_openpi_blog","src_openpi_repo"]}},"commercialAvailability":{"value":"orderable","asOf":"2026-10-18","sourceIds":["src_openpi_repo"],"note":"Free to download. Google's Gemma Terms of Use, included in the repo, apply to the models."},"lifecycle":{"value":"active","asOf":"2026-10-18","sourceIds":["src_openpi_repo"],"note":"Repository last updated on August 24, 2026."},"releaseChannel":"general","statusLabel":"Free open models","availability":"Free to download. The code is Apache 2.0; the models are built on Google's PaliGemma and come with Google's Gemma Terms of Use.","pricing":[{"label":"Download","amount":0,"currency":"USD","cadence":"one_time","note":"Free. Google's Gemma Terms of Use apply to the models.","checkedAt":"2026-10-18","sourceId":"src_openpi_repo"}],"claims":[{"id":"clm_openpi_release","text":"Physical Intelligence released the weights and code for π0 and its π0-FAST model on February 4, 2025, and π0.5, which it describes as an upgraded π0 with better open-world generalization, in September 2025. The repository lists base models for fine-tuning and versions fine-tuned for the DROID, ALOHA and LIBERO robot setups.","evidenceClass":"documented","sourceIds":["src_pi_openpi_blog","src_openpi_repo"],"observedAt":"2026-10-18"},{"id":"clm_openpi_license","text":"The openpi code is licensed under Apache 2.0. The repository also includes Google's Gemma Terms of Use, and its models are built on Google's PaliGemma; Gemma's terms cover models derived from Gemma, including its prohibited-use policy and conditions on redistribution. The repository does not name a separate licence for the weights.","evidenceClass":"documented","sourceIds":["src_openpi_repo"],"observedAt":"2026-10-18"},{"id":"clm_openpi_phail","text":"In the PhAIL real-robot benchmark (May 2026), π0.5 and NVIDIA's GR00T N1.6 tied for best among four open models on a bin-picking task (13.8% and 13.3% of a human operator's throughput, statistically indistinguishable); the best model was roughly seven times slower per operation than a person. The author's company runs the benchmark, and annotation was not blinded.","evidenceClass":"documented","sourceIds":["src_phail_benchmark"],"observedAt":"2026-10-18","support":{"src_phail_benchmark":{"support":"independently_verified","verified":"PhAIL fine-tuned π0.5, GR00T N1.6, ACT and SmolVLA on the same 449 demonstrations and tested them on a real Franka arm picking four kinds of objects; the author's company, Positronic Robotics, runs the benchmark, and annotation was not blinded"}}},{"id":"clm_openpi_arena","text":"NVIDIA's Isaac Lab Arena documents running openpi's π0 and π0.5 DROID checkpoints in simulation through openpi's policy server, and notes that \"VLA models are improving but not yet fully robust under zero-shot distribution shift.\"","evidenceClass":"documented","sourceIds":["src_isaaclab_arena_openpi"],"observedAt":"2026-10-18","support":{"src_isaaclab_arena_openpi":{"support":"independently_verified","verified":"NVIDIA's Isaac Lab Arena documentation shows how to run openpi's π0 and π0.5 DROID checkpoints against its simulation through openpi's policy server"}}}],"sourceIds":["src_pi_openpi_blog","src_openpi_repo","src_phail_benchmark","src_isaaclab_arena_openpi"],"lastCheckedAt":"2026-10-18","updatedAt":"2026-10-18","editorial":{"status":"published","reviewedBy":"Holly White","reviewedAt":"2026-10-18","note":"Batch 2B-F record, approved in the evidence review (2026-10-18) and published."},"status":"available","statusAsOf":"2026-10-18","sourceRecords":[{"id":"src_pi_openpi_blog","title":"Open Sourcing π0","publisher":"Physical Intelligence","url":"https://www.pi.website/blog/openpi","origin":"company","mediaType":"official_page","publishedAt":"2025-02-04","checkedAt":"2026-10-18","sourceType":"first_party"},{"id":"src_openpi_repo","title":"Physical-Intelligence/openpi","publisher":"Physical Intelligence","url":"https://github.com/Physical-Intelligence/openpi","origin":"company","mediaType":"code_repository","checkedAt":"2026-10-18","sourceType":"first_party"},{"id":"src_phail_benchmark","title":"PhAIL: A Real-Robot VLA Benchmark and Distributional Methodology","publisher":"arXiv (Positronic Robotics)","url":"https://arxiv.org/abs/2605.29710","origin":"independent_reporting","mediaType":"independent_test","publishedAt":"2026-05-28","checkedAt":"2026-10-18","sourceType":"independent"},{"id":"src_isaaclab_arena_openpi","title":"OpenPI — Isaac Lab Arena","publisher":"NVIDIA","url":"https://isaac-sim.github.io/IsaacLab-Arena/main/pages/quickstart/running_a_real_policy/openpi.html","origin":"independent_reporting","mediaType":"documentation","checkedAt":"2026-10-18","sourceType":"independent"}],"canonicalUrl":"https://itsalive.ai/products/openpi"}