BitNet b1.58
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
- University of Chinese Academy of Sciences,Microsoft Research
- Organisation type
- Academia,Industry
- Country
- China, United States of America
- Published
- 27 February 2024
- Authors
- Shuming Ma, Hongyu Wang, Lingxiao Ma, Lei Wang, Wenhui Wang, Shaohan Huang, Li Dong, Ruiping Wang, Jilong Xue, Furu Wei
What it does
The problem areas the model was built for. A model can carry several of each.
- Domain
- Language
- Task
- Language modeling/generation, 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
- 70B
- Training data
- 100,000,000,000 tokens
"we pre-trained the models on the RedPajama dataset [Com23] for 100 billion tokens."
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
- 2.9 × 10²² FLOP
- How it was established
- Operation counting,Comparison with other models
6ND = 6*70*10^9*100*10^9 = 42000000000000000000000 (4.2e+22) Figure 3 suggests it 41.2 times more energy efficient than LLAMA 70B (which is estimated 8.1e+23 FLOPS -> 1.9660194e+22 FLOPs geometric mean -> 2.8735486e+22
How it is classified
Labels the source dataset applies when tracking notable models, and how confident it is in the entry.
- Record confidence
- Confident
Sources
Where this record came from and when it was last checked.
- Reference
- The Era of 1-bit LLMs: All Large Language Models are in 1.58 Bits
- Last updated
- 28 November 2025
What the numbers mean
What this model is
BitNet b1.58 was published by University of Chinese Academy of Sciences,Microsoft Research, in the country recorded as China, during February 2024. It comes out of an organisation categorised as academia,Industry.
It works in the domain of Language, and is recorded as performing the task of language modeling/generation, Question answering.
Because the weights are not available, none of the hardware figures elsewhere on this site apply to it.
Training and provenance
Training it took a computation budget of roughly 2.9 × 10²² FLOP. That figure measures what producing the model cost, and has no bearing on how fast it answers.
It was trained on a corpus of about 100,000,000,000 tokens of text.
Answers
BitNet b1.58 — common questions
BitNet b1.58— 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.
BitNet b1.58— is it open source?
The licensing was never recorded in our source data. We treat unstated licensing as closed, because an unrecorded licence is not one to rely on.
BitNet b1.58— how many parameters does it have?
It has a parameter count of 70B. 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.
BitNet b1.58— who created it?
It was published by University of Chinese Academy of Sciences,Microsoft Research, based in China, an organisation categorised as academia,Industry.
BitNet b1.58— when was it released?
It was published in February 2024. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.
BitNet b1.58— what is it used for?
It works in the domain of Language, and is recorded as handling the task of language modeling/generation, Question answering. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.
BitNet b1.58— how much compute was used to train it?
Training consumed around 2.9 × 10²² 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.
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.