Skip to content
InvestingStack
ModelsEnterprise

BloombergGPT

The 50-billion-parameter proprietary model that started the financial-LLM race

By Bloomberg

Visit websitePaper · arXiv 2303.17564
BloombergGPT is not open source, so there are no GitHub metrics to track. We list it for completeness and describe only what can be verified from public documentation.

What it is

BloombergGPT was trained on a mixed corpus of Bloomberg's financial archive and public data. It is not available outside Bloomberg; it matters as the proof point that domain-specific pre-training helps on financial NLP tasks, and as the benchmark that FinGPT and PIXIU set out to match with open weights.

Tags

  • Proprietary
  • Paper
  • Domain pre-training