Physical Intelligence · robot AI model
π0.6
United States · 2025
Sources retrieved: 10 November 2026 (A, B, C)
- Model type
- Vision-language-action (VLA)Source and note: Model typeA website.pi-asset.com/pi06star/PI06_model_card.pdf · Retrieved 10 November 2026Note. Developer wording: π0.6 is “our newest vision-language-action (VLA) model that builds on top of π0.5”.
- What it controls
- Arms and handsSource and note: What it controlsB pi.website/download/pistar06.pdf · Retrieved 10 November 2026Note. π*0.6 paper: the model produces action chunks “which consists of joint angles and gripper commands at 50 Hz”.
- Inputs
- Camera Language Robot stateSource and note: InputsA website.pi-asset.com/pi06star/PI06_model_card.pdf · Retrieved 10 November 2026Note. Model card: during pre-training “the model receives up to four images as input”, “corresponding to a base camera, up to two wrist cameras, and an optional backward camera for mobile manipulators”; the image tokens are “concatenated with the tokenized language prompt and tokenized proprioceptive states”.
- Base model
- Gemma3 4BSource and note: Base modelA website.pi-asset.com/pi06star/PI06_model_card.pdf · Retrieved 10 November 2026Note. Model card: “The vision-language backbone is initialized from the Gemma3 4B model”. π*0.6 is built on π0.6: “π * 0.6 is based on the π 0.6 model”.
Specifications
7 of 14 stated 7 not stated
4 fields hidden 3 not-stated fields with a note stay ·
How to read the figuresShow guideClose
- not statedNo source
- yesStated by a source
- 0Reported as zero
- ASource
Select a source marker for source details and notes.
Release1/1 disclosed
Released17 November 2025
Source and note: ReleasedA website.pi-asset.com/pi06star/PI06_model_card.pdf · Retrieved 10 November 2026Note. “November 17, 2025”
Input and output3/4 disclosed
Model typeVision-language-action (VLA)
Source and note: Model typeA website.pi-asset.com/pi06star/PI06_model_card.pdf · Retrieved 10 November 2026Note. Developer wording: π0.6 is “our newest vision-language-action (VLA) model that builds on top of π0.5”.
InputsCamera Language Robot state
Source and note: InputsA website.pi-asset.com/pi06star/PI06_model_card.pdf · Retrieved 10 November 2026Note. Model card: during pre-training “the model receives up to four images as input”, “corresponding to a base camera, up to two wrist cameras, and an optional backward camera for mobile manipulators”; the image tokens are “concatenated with the tokenized language prompt and tokenized proprioceptive states”.
What it controlsArms and hands
Source and note: What it controlsB pi.website/download/pistar06.pdf · Retrieved 10 November 2026Note. π*0.6 paper: the model produces action chunks “which consists of joint angles and gripper commands at 50 Hz”.
Degrees of freedom controlledNot stated
Source and note: Degrees of freedom controlledB pi.website/download/pistar06.pdf · Retrieved 10 November 2026Note. The π*0.6 paper uses “a static bimanual system with two 6 DoF arms with parallel jaw grippers”.
Size and speed1/2 disclosed
ParametersNot stated
Source and note: ParametersA website.pi-asset.com/pi06star/PI06_model_card.pdf · Retrieved 10 November 2026Note. Model card: the backbone is “SigLIP (400M) + Gemma (4B)”, and the action expert “consists of about 860M parameters”. Blog post: “The π 0.6 VLA is based on a 5B-parameter vision-language model, augmented with an action expert.”
Control rate50Hz
Source and note: Control rateB pi.website/download/pistar06.pdf · Retrieved 10 November 2026Note. Action execution layer, the fastest stated, not inference: the model produces action chunks “which consists of joint angles and gripper commands at 50 Hz”; on the test robot “The arms are controlled at 50 Hz with joint positions.” Model card: π0.6 “takes 63ms to produce an action chunk on a single H100 GPU”.
Where it runs and what it is built on1/2 disclosed
On robot or cloudNot stated
Base modelGemma3 4B
Source and note: Base modelA website.pi-asset.com/pi06star/PI06_model_card.pdf · Retrieved 10 November 2026Note. Model card: “The vision-language backbone is initialized from the Gemma3 4B model”. π*0.6 is built on π0.6: “π * 0.6 is based on the π 0.6 model”.
Openness0/3 disclosed
AccessNot stated
Licence documentNot stated
Licence typeNot stated
The developer’s own claims1/2 disclosed
Learns new tasksNot stated
Source and note: Learns new tasksC pi.website/blog/pistar06 · Retrieved 10 November 2026Note. Physical Intelligence states this for π*0.6 only: “we fine-tune π * 0.6 to each task with demonstrations, and then train it further with RL using additional data collected on the robot”. The π0.6 model card reports results without task-specific training: “These experiments evaluate each model out of the box, without finetuning.”
Works across robotsYes
Source and note: Works across robotsA website.pi-asset.com/pi06star/PI06_model_card.pdf · Retrieved 10 November 2026Note. Developer claim: the model card evaluates π0.6 on tasks “performed with static robot platforms” and on tasks “performed with mobile bimanual robots”; its training data include “cross-embodiment data collected in-house”.Yes only if the developer itself says the model runs on different robot models, for example robots from two manufacturers or two of one manufacturer’s models. Several units of the same robot do not count.
Versions
- π0.6
- π*0.6
The developer has released this model in several versions. Each figure applies to the version its source names.
Notes
Physical Intelligence introduces π0.6 as “our newest vision-language-action (VLA) model that builds on top of π0.5”.
π*0.6 is π0.6 further trained with Recap “(RL with Experience & Corrections via Advantage-conditioned Policies)”, a method of learning from demonstrations, corrections and the robot's own experience. Physical Intelligence: “We call the resulting RL-trained model π * 0.6”.
Physical Intelligence published the π0.6 model card and the π*0.6 blog post on 17 November 2025.
Country and city from the business database CB Insights (396 Treat Avenue, San Francisco); the developer does not state its location.
Sources
One line per source, not per figure: a source has one retrieval date.
- Awebsite.pi-asset.com/pi06star/PI06_model_card.pdfMain source. Retrieved 10 November 2026
- Bpi.website/download/pistar06.pdfRetrieved 10 November 2026
- Cpi.website/blog/pistar06Retrieved 10 November 2026
