University of California, BerkeleyUnited States2024

Octo

An AI model: a vision-language-action model (VLA) takes camera images and an instruction and turns them into movements for a robot. This page describes the software, not a robot.

Robotsnot stated

VersionsOcto-SmallOcto-Base

The developer has released this model in several versions. Each figure applies to the version its source names.

Key figures as stated by the developer

5 of 6 stated · 1 gap

Model type not stated
Controls ⁠A arms and hands
Inputs ⁠A camera language
Access ⁠A open download
Parameters ⁠C 93,000,000
Base model ⁠B T5-Base language encoder (language only)

How to read the figures

14kgA30.9lbA: The letter in the grey square after a figure points to the source the figure came from. Robots with several sources carry several letters.
74.5lb: Imperial figures are our own conversion unless the manufacturer states an imperial figure itself. The source letter always belongs to the metric figure — a conversion has no source of its own.

The four states

Versions and notes in the key figures (6)

Model type

Note The developer calls it something else and does not label it vision-language-action or world action model: "Octo models are transformer-based diffusion policies".

Controls

Developer: it "can control various robot arms".

Inputs

Developer: "Out of the box, Octo supports multiple RGB camera inputs, can control various robot arms, and can be instructed via language commands or goal images." Robot state only after fine-tuning: Octo is designed for "efficient finetuning to new robot setups, including robots with different actions and different combinations of cameras and proprioceptive information". Goal images have no value of their own in the closed list.

Access

README: "You can find pretrained Octo checkpoints" on Hugging Face.

Parameters

Two sizes: "Octo-Small (27M parameters) and Octo-Base (93M parameters)". Octo-Base is shown; Octo-Small is a variant.

Versions

Octo-Small: 27,000,000
Octo-Base: 93,000,000

Base model

Model card: "Language is tokenized by applying the T5 tokenizer, and then applying the T5-Base language encoder."

Full specifications

8 of 12 fields stated

The letter points to the source at the foot of the page. Fields without a figure are dashed: the developer does not state them — that is not the same as zero, and not the same as no.
Field	Value	Source and caveats
Input and output
Model type	not stated	Note The developer calls it something else and does not label it vision-language-action or world action model: "Octo models are transformer-based diffusion policies". ⁠A
Inputs	`camera` `language`	A Developer: "Out of the box, Octo supports multiple RGB camera inputs, can control various robot arms, and can be instructed via language commands or goal images." Robot state only after fine-tuning: Octo is designed for "efficient finetuning to new robot setups, including robots with different actions and different combinations of cameras and proprioceptive information". Goal images have no value of their own in the closed list.
Controls	`arms and hands`	A Developer: it "can control various robot arms".
Degrees of freedom controlled	not stated	Note Octo-Base is described as "predicting 7-dimensional actions 4 steps into the future using a diffusion policy"; the number of degrees of freedom controlled is not stated. ⁠B
Size and speed
Parameters	93,000,000	C Two sizes: "Octo-Small (27M parameters) and Octo-Base (93M parameters)". Octo-Base is shown; Octo-Small is a variant. Versions Octo-Small 27,000,000 Octo-Base 93,000,000
Control rate	not stated
Where it runs and what it is built on
Runs where	not stated
Base model	T5-Base language encoder (language only)	B Model card: "Language is tokenized by applying the T5 tokenizer, and then applying the T5-Base language encoder."
Openness
Access	open download	A README: "You can find pretrained Octo checkpoints" on Hugging Face.
License	MIT	B Model card: "license: mit".
The developer’s own claims
Learns new tasks	needs retraining	C Developer claim: the project page tests Octo's ability "to efficiently finetune to new tasks and environments with small target domain datasets".
Works across robots	yes	A Developer claim: it "can control various robot arms". 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.

Notes

The Octo repository describes Octo models as transformer-based diffusion policies trained on a diverse mix of 800k robot trajectories.
The Octo paper was first submitted to arXiv on 20 May 2024.
The UC Berkeley admissions page gives its address as Berkeley, CA.

Sources

A Manufacturer product page. github.com/octo-models/octo · Retrieved 2026-10-13
B huggingface.co/rail-berkeley/octo-base-1.5 · Retrieved 2026-10-13
C octo-models.github.io/ · Retrieved 2026-10-13

One line per source, not per figure: a source has one retrieval date. The letter beside a figure points to the line it came from.