
Pi0.5
Open vision-language-action model for robots that generalizes to new environments.
About
Pi0.5 (pi0.5) is Physical Intelligence's vision-language-action (VLA) foundation model for generalist robot control, published in April 2025 as arXiv 2504.16054 (exact release day not confirmed). It upgrades the earlier pi0 with knowledge insulation, a technique aimed at the core robotics challenge of open-world generalization: performing tasks correctly in new settings with new objects, not just controlled lab environments.
Like the rest of the openpi family, pi0.5 shares a PaliGemma backbone, is pre-trained on 10,000+ hours of multi-platform robot data, and is released under Apache 2.0 with JAX and PyTorch implementations. Base checkpoints can be fine-tuned on custom tasks, with the PyTorch path workable on a single GPU, and LeRobot ships an adapted implementation (lerobot/pi05_base on Hugging Face).
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 94; Trending score 84/100 from researcher assessment.