Wu Dao 2.0
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
- Beijing Academy of Artificial Intelligence / BAAI
- Organisation type
- Academia
- Country
- China
- Published
- 31 May 2021
- Authors
- Tang Jie, Zhai Jidong, Yang Hongxia, Chen Wenguang, Zheng Weimin, Ma Zixuan, He Jiaao, Qiu Jiezhong, Cao Huanqi, Wang Yuanwei, Sun Zhenbo, Zheng Liyan, Wang Haojie, Tang Shizhi, Feng Guanyu, Zeng Aohan, Zhong Runxin, Shi Tianhui, Du Zhengxiao, Ding Ming, Tiago Antunes, Peng Jinjun, Lin Junyang Zhang Jianwei
What it does
The problem areas the model was built for. A model can carry several of each.
- Domain
- Multimodal, Language, Vision, Image generation
- Task
- Image captioning, Chat, Image generation, Text-to-image, Language modeling/generation, Question answering, Visual question answering
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
- 1.8T
- Training data
- 4,900,000,000,000 tokens
"It's been trained on 1.75 trillion parameters" MoE architecture, "tens of thousands of experts" https://keg.cs.tsinghua.edu.cn/jietang/publications/wudao-3.0-meta-en.pdf
[tokens assumed[ "Bilingual (Cn and En) data: 4.9T text and images" https://keg.cs.tsinghua.edu.cn/jietang/publications/wudao-3.0-meta-en.pdf
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
- 1.5 × 10²⁴ FLOP
- How it was established
- Operation counting
It's a mixture-of-experts model, so all 1.75 trillion params were most likely not trained on each token. "The parameter scale of Enlightenment 2.0 reached a record-breaking 1.75 trillion. According to reports, the new generation FastMoE technology is the key to the realization of the "Trillion Model" cornerstone of Enlightenment 2.0." Speculatively assuming 3% of parameters are active per forward pass and the model was trained for one epoch: 6 FLOP / token / parameter * 1.75 * 10^12 parameter…
The training run
What it physically took to train: which chips, how many, for how long, and what that drew from the wall.
- Compute cost
- $2,870,840
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
- API access
How it is classified
Labels the source dataset applies when tracking notable models, and how confident it is in the entry.
- Frontier model
- Yes
- Foundation model
- Yes
- Record confidence
- Speculative
Sources
Where this record came from and when it was last checked.
- Reference
- China's gigantic multi-modal AI is no one-trick pony
- Last updated
- 1 December 2025
What the numbers mean
Background
Wu Dao 2.0 was published by Beijing Academy of Artificial Intelligence / BAAI, in the country recorded as China, during May 2021. The publishing organisation is categorised as academia.
It works in the domain of Multimodal, Language, Vision, Image generation, and is recorded as performing the task of image captioning, Chat, Image generation, Text-to-image, Language modeling/generation, Question answering, Visual question answering.
This is a closed model: the trained values stayed with whoever produced them, and there is no local version to run.
What went into building it
Training it took a computation budget of roughly 1.5 × 10²⁴ FLOP. That figure measures what producing the model cost, and has no bearing on how fast it answers.
Training consumed a corpus of around 4,900,000,000,000 tokens of text.
Answers
Wu Dao 2.0 — common questions
Wu Dao 2.0— when was it released?
It was published in May 2021. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.
Wu Dao 2.0— what is it used for?
It works in the domain of Multimodal, Language, Vision, Image generation, and is recorded as handling the task of image captioning, Chat, Image generation, Text-to-image, Language modeling/generation, Question answering, Visual question answering. These are the areas it was designed around; they describe intent rather than a hard boundary.
Wu Dao 2.0— how much compute was used to train it?
Training consumed around 1.5 × 10²⁴ FLOP. 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.
Wu Dao 2.0— what GPU do I need to run it?
None. This 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.
Wu Dao 2.0— is it open source?
No. Its weights have not been published, so it exists only as a service controlled by its owner.
Wu Dao 2.0— how many parameters does it have?
It has a parameter count of 1.8T. "It's been trained on 1.75 trillion parameters" MoE architecture, "tens of thousands of experts" https://keg.cs.tsinghua.edu.cn/jietang/publications/wudao-3.0-meta-en.pdf. 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.
Wu Dao 2.0— who created it?
It was published by Beijing Academy of Artificial Intelligence / BAAI, based in China, an organisation categorised as academia.
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.