Physical Intelligence

π0.5 (per-task)↗

π0.5 (per-task) is a VLA from Physical Intelligence, released April 22, 2025. It has no overall rank on the Robot Intelligence Index, which needs both a simulation and a real-robot result. Its best result is #5 of 22 on RoboChallenge. π0 successor co-trained on heterogeneous web and robot data for open-world generalization. (Fine-tuned separately per task.)

Release date: Apr 22, 2025

Type
VLA
Model weights
Open
License
Apache-2.0
Parameters
3.6B
Pricing
Self-hosted
Benchmarks
1
Overall rank
Not ranked
Needs a simulation result to rank overall.
Real robots rank
#18of 56
Rating 1619
Simulation rank
Not ranked
No simulation results yet.
Rating
1601
10601989
Cost / task · est.
$0.255
$0.0087$17.74
Time / task
4min39s
9.5 s1 h 22 min
Rank on each benchmark
BenchmarkScoreRank
RoboChallengeProgress61.85 / 22

Per task attempt; the tiles show the median across benchmarks. Self-hosted compute is priced at one H100 at $3.29 per hour (Lambda on-demand, Oct 2026).

Other models from Physical Intelligence

Inference

Independent benchmark (openpi JAX defaults, batch 1): 76.1 ms per 15-action chunk on RTX 4090; 44.2 ms H100; 49.3 ms H200; 53.9 ms RTX PRO 6000; 35.9 ms B300. Params 3.6B per lerobot/pi05_base safetensors.