BloombergGPT
A 50 billion parameter language model for finance.
BloombergGPT is a large language model with 50 billion parameters, designed specifically for financial tasks. Developed by Bloomberg and released in 2023, it trains on a mix of general and extensive proprietary financial data to outperform other models in its domain while maintaining general capabilities.
- Contains 50 billion parameters.
- Trained on 363 billion financial tokens.
- Released in March 2023 by Bloomberg.
- Outperforms existing models on financial tasks.
- Maintains strong general language benchmark scores.
What it is
BloombergGPT is a large language model built for the financial sector. It contains 50 billion parameters and was introduced in March 2023 by Bloomberg. The model addresses the lack of specialised tools in financial technology, handling tasks such as sentiment analysis, named entity recognition, and question answering. Unlike general models, it is trained on a vast collection of financial data alongside standard text sources. This approach allows it to perform better on industry-specific benchmarks without losing competence in general language tasks.
How it came about
The model was presented in a paper submitted to arXiv in March 2023. The work was carried out by a team including Shijie Wu, Ozan Irsoy, and Mark Dredze among others. Prior to this release, no large language model dedicated to the financial domain had been reported in academic literature. The authors updated their publication in December 2023 to include detailed training chronicles, sharing their experience in building the system.
Inside the model
The architecture relies on a mixed dataset strategy. The team constructed a corpus of 363 billion tokens from Bloomberg’s extensive internal data sources. They augmented this with 345 billion tokens from general purpose datasets. This combination totals 708 billion tokens. The training process aimed to balance domain expertise with general linguistic ability. The developers documented their modelling choices and evaluation methods to explain how the system achieves high performance on both financial and standard benchmarks.
What it can do
BloombergGPT is validated against standard large language model benchmarks, open financial benchmarks, and internal tests that reflect intended usage. The results show that the mixed dataset training enables the model to outperform existing systems on financial tasks by significant margins. Crucially, this specialisation does not sacrifice performance on general language benchmarks. Applications include complex natural language processing tasks like sentiment analysis and question answering within the financial context.
How to use it
The model weights are not openly available for public download. Access is primarily through the research community via the published paper and associated documentation. The authors released Training Chronicles in an appendix to detail their experience.
Why it matters
This work fills a gap in the literature by presenting the first large language model specialised for finance. It demonstrates that domain-specific data can significantly boost performance on industry tasks without degrading general capabilities. The release of training details provides a reference for future developments in specialised natural language processing. It sets a precedent for using large proprietary datasets to enhance model utility in specific professional fields.
Questions people ask about BloombergGPT
What is BloombergGPT used for?
It is designed for financial technology applications. These include sentiment analysis, named entity recognition, and question answering. It helps process complex financial data using natural language processing techniques.
How many parameters does the model have?
BloombergGPT has 50 billion parameters. This size allows it to handle complex linguistic patterns while remaining efficient enough to be trained on a highly specialised dataset.
Is the model open source?
No, the weights are not publicly released. The developers shared their methodology and training chronicles in a research paper, but the model itself is not available for open download or community modification.
Good reading and sources
There is no English Wikipedia article on BloombergGPT yet, so the facts on this page come from the sources listed above: arxiv.org/abs/2303.17564. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.