Large language models (LLMs) have reshaped AI research, but their development has been largely confined to well-funded industrial labs. Today, a global collaboration of over 1,000 researchers from 70+ countries and 250+ institutions releases BLOOM, the first multilingual LLM built with complete transparency. With 176 billion parameters, BLOOM can generate text in 46 natural languages and 13 programming languages, making it the largest open model for many languages including Spanish, French, and Arabic.
Training BLOOM took 117 days on the Jean Zay supercomputer in France, supported by a €3 million compute grant from CNRS and GENCI. The model is now freely available for download, study, and use under a Responsible AI License. For those without expensive hardware, an inference API backed by Google's TPU cloud enables easy experimentation.
BLOOM is not a one-off release; it is the foundation for a growing family of models. Future work includes making it instruction-tuned, adding more languages, and creating compressed versions for broader accessibility. “All the experiments researchers have always wanted to run, starting with the power of a 100+ billion parameter model, are now possible,” the team says.