ViT-22B
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
- Organisation type
- Industry
- Country
- United States of America
- Published
- 10 February 2023
- Authors
- Mostafa Dehghani, Josip Djolonga, Basil Mustafa, Piotr Padlewski, Jonathan Heek, Justin Gilmer, Andreas Steiner, Mathilde Caron, Robert Geirhos, Ibrahim Alabdulmohsin, Rodolphe Jenatton, Lucas Beyer, Michael Tschannen, Anurag Arnab, Xiao Wang, Carlos Riquelme, Matthias Minderer, Joan Puigcerver, Utku Evci, Manoj Kumar, Sjoerd van Steenkiste, Gamaleldin F. Elsayed, Aravindh Mahendran, Fisher Yu, Av…
What it does
The problem areas the model was built for. A model can carry several of each.
- Domain
- Vision
- Task
- Object detection, Image classification
- Numerical format
- BF16
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
- 21.7B
- Training data
- 4,000,000,000 tokens
- Epochs
- 2.9
21.743B, Table 1
"Dataset. ViT-22B is trained on a version of JFT (Sun et al., 2017), extended to around 4B images (Zhai et al., 2022a). These images have been semi-automatically annotated with a class-hierarchy of 30k labels"
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
- 1.9 × 10²³ FLOP
- How it was established
- Hardware
"ViT-22B was trained using 256 visual tokens per image, where each token represents a 14 × 14 patch extracted from 224 × 224 sized images. ViT-22B is trained for 177k steps with batch size of 65k: approximately 3 epochs" "ViT-22B was trained on 1024 TPU V4 chips for 177K steps" "Using these techniques, ViT-22B processes 1.15k tokens per second per core during training (forward and backward pass) on TPUv4 [...] ViT-22B’s model flops utilization (MFU) is 54.9%" 256 * 177k * 65k = 2.945T tokens …
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
- Wall-clock time
- 347 hours (14.5 days)
- Hardware utilisation
- MFU 54.9%
- Power draw
- 694.9 kW
- Compute cost
- $285,556
"Using these techniques, ViT-22B processes 1.15k tokens per second per core during training (forward and backward pass)" From model card we know they trained with 1024 TPUv4 chips, and there are 2 cores per chip. Total number of tokens was 177K steps * 65k images/step * 256 tokens/image = 2.945T tokens So training time is 2.945T tokens / (1.15k * 2 * 1024) tokens/s = 1.25M seconds = 347.4 hours
"Using these techniques, ViT-22B processes 1.15k tokens per second per core during training (forward and backward pass) on TPUv4 [...] ViT-22B’s model flops utilization (MFU) is 54.9%"
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
don't see it here: https://github.com/google-research/vision_transformer?tab=readme-ov-file#available-vit-models
How it is classified
Labels the source dataset applies when tracking notable models, and how confident it is in the entry.
- Likely above 10²³ FLOP
- Yes
- Why it is tracked
- SOTA improvement
- Record confidence
- Confident
- Citations
- 862
"The largest ViT-22B sets the new SOTA on the challenging ObjectNet test set"
Sources
Where this record came from and when it was last checked.
- Reference
- Scaling Vision Transformers to 22 Billion Parameters
- Last updated
- 25 May 2026
What the numbers mean
Where it came from
ViT-22B was published by Google, in United States of America, in February 2023. industry is the category the publisher falls under.
It works in Vision, and is recorded as doing object detection, Image classification.
Its weights were never published, so it can only be reached through its provider. No graphics card changes that.
What went into building it
The training run consumed about 1.9 × 10²³ FLOP, on Google TPU v4. That figure describes the cost of creating it and has no bearing on how quickly it generates text.
It was trained on about 4,000,000,000 tokens of text.
It is tracked in the underlying dataset for one reason in particular: sOTA improvement.
Answers
ViT-22B — common questions
How much compute was used to train ViT-22B?
Around 1.9 × 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.
What GPU do I need to run ViT-22B?
None. ViT-22B 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.
Is ViT-22B open source?
No. ViT-22B has not had its weights published, so it exists only as a service controlled by its owner.
How many parameters does ViT-22B have?
ViT-22B has 21.7B parameters. 21.743B, Table 1. 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.
Who created ViT-22B?
ViT-22B was published by Google, based in United States of America, categorised as industry.
When was ViT-22B released?
ViT-22B was published in February 2023. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.
What is ViT-22B used for?
ViT-22B works in Vision, and is recorded as handling object detection, Image classification. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.
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.