PaLI

Closed weights Google 16.9B parameters September 2022

No estimate

No hardware requirements for this model

The weights for this model have not been published, so it cannot be downloaded or run on your own hardware at any size. It is reachable only through its provider, and no graphics card changes that.

On record

Full specification

Everything on record for this model. Most of it describes how it was trained rather than how it runs — useful context for judging how much work went into it, and how it compares with models built at a different scale.

Origin

Who built this model, where, and when it was published.

Organisation
Google
Organisation type
Industry
Country
United States of America
Published
14 September 2022
Authors
Xi Chen, Xiao Wang, Soravit Changpinyo, AJ Piergiovanni, Piotr Padlewski, Daniel Salz, Sebastian Goodman, Adam Grycner, Basil Mustafa, Lucas Beyer, Alexander Kolesnikov, Joan Puigcerver, Nan Ding, Keran Rong, Hassan Akbari, Gaurav Mishra, Linting Xue, Ashish Thapliyal, James Bradbury, Weicheng Kuo, Mojtaba Seyedhosseini, Chao Jia, Burcu Karagol Ayan, Carlos Riquelme, Andreas Steiner, Anelia Angelo…

What it does

The problem areas the model was built for. A model can carry several of each.

Domain
Language, Vision, Multimodal
Task
Visual question answering, Language modeling/generation, Image captioning

Size

How large the model is and how much data it was trained on. Parameters are the figure that decides whether it fits on a given graphics card.

Parameters
16.9B

3.9b Image Encoder, 14b Multimodal Encoder-Decoder

Training data
143,507,000,000 tokens

"During training, the model passes over 1.6B images, one epoch over the entire pretraining dataset"

Epochs
1

Training compute

The arithmetic performed to train the model, measured in floating-point operations. It is a measure of what the training run cost, not of how fast the finished model answers you.

Training compute
1.7 × 10²³ FLOP

Pre-training the ViT component involved 1.1 million steps (they train over 1M steps but run the last 100k twice and then average the two resulting models). Batch size is 16384 and the inputs are 224x224. Table 8 indicates a forward pass with ViT-e/14 on a 224 image takes 1980 GFLOPs, so total training compute for the ViT-e/14 model is: 1980e9 * 16384 * 1.1 million * 3 (account for backward passes) = 1.07e23 In the "overal model" section, they then say: "The largest model, PaLI-17B, is pretraine…

How it was established
Operation counting,Hardware

The training run

What it physically took to train: which chips, how many, for how long, and what that drew from the wall.

Training hardware
Google TPU v4
Chips used
1,024
Chip-hours
172,032
Wall-clock time
240 hours (10 days)

10

Power draw
697.2 kW
Compute cost
$50,878

Availability

Whether you can obtain the model and run it on your own hardware, which is what decides if any of the graphics-card figures on this page apply.

Weights
Closed — provider access only
Model access
Unreleased
Training code
Unreleased

How it is classified

Labels the source dataset applies when tracking notable models, and how confident it is in the entry.

Foundation model
Yes
Likely above 10²³ FLOP
Yes
Why it is tracked
SOTA improvement

"PaLI achieves state-of-the-art in multiple vision and language tasks (such as captioning, visual question-answering, scene-text understanding)" "PaLI-17B achieves state-of-the-art (SOTA) results on multiple benchmarks, outperforming some strong models. Specifically, PaLI outperforms recent and concurrent models on the long-standing COCO Captioning benchmark (Chen et al., 2015), with 149.1 CIDEr score on the Karpathy split (Karpathy & Fei-Fei, 2015). PaLI also achieves a new SOTA of 84.3% on VQ…

Record confidence
Likely
Citations
983

Sources

Where this record came from and when it was last checked.

Reference
PaLI: A Jointly-Scaled Multilingual Language-Image Model
Last updated
25 May 2026

What the numbers mean

Background

PaLI was published by Google, in United States of America, in September 2022. The organisation is categorised as industry.

It works in Language, Vision, Multimodal, and is recorded as doing visual question answering, Language modeling/generation, Image captioning.

This is a closed model: the trained values stayed with whoever produced them, and there is no local version to run.

Training and provenance

Training it took roughly 1.7 × 10²³ FLOP of computation, on Google TPU v4 — a measure of what producing the model cost, not of how fast it answers.

It was trained on about 143,507,000,000 tokens of text.

Its inclusion criterion is sOTA improvement.

Answers

PaLI — common questions

01

How much compute was used to train PaLI?

Around 1.7 × 10²³ FLOP, on Google TPU v4. That measures what producing the model cost and says nothing about how quickly it answers once trained — inference speed comes from memory bandwidth, not from the training budget.

02

What GPU do I need to run PaLI?

None. PaLI is a closed model — its weights were never published, so it cannot be downloaded or run on your own hardware at any price. It is reachable only through its provider.

03

Is PaLI open source?

No. PaLI has not had its weights published, so it exists only as a service controlled by its owner.

04

How many parameters does PaLI have?

PaLI has 16.9B parameters. 3.9b Image Encoder, 14b Multimodal Encoder-Decoder. That figure is the total, and it is what decides how much memory the model needs — roughly half a gigabyte per billion at the compression most people use.

05

Who created PaLI?

PaLI was published by Google, based in United States of America, categorised as industry.

06

When was PaLI released?

PaLI was published in September 2022. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.

07

What is PaLI used for?

PaLI works in Language, Vision, Multimodal, and is recorded as handling visual question answering, Language modeling/generation, Image captioning. A model can carry several of each, so these are the areas it was built for rather than a limit on what it will attempt.

Source

Original publication

Record last updated 25 May 2026

The other direction

Looking at it from the other side?

This page starts from the model. If you already own a card and want to know everything it will run, start from the hardware instead.