BLOOM
BigScience's 2022 open multilingual model, still downloadable, now a research artefact
Overview
BLOOM is a 176-billion-parameter multilingual language model released in July 2022 by BigScience, an open research workshop of more than 1,000 researchers coordinated by Hugging Face. It covers 46 natural and 13 programming languages, was trained on France's Jean Zay supercomputer, and remains downloadable from Hugging Face under the BigScience RAIL License v1.0 along with smaller sizes and the BLOOMZ fine-tunes.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 14 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
Use cases
What you can produce with BLOOM
- Downloadable 176B multilingual weights covering 46 natural and 13 programming languages
- Smaller BLOOM sizes from 560M to 7.1B parameters for modest hardware
- BLOOMZ instruction-tuned variants for zero-shot prompting
- Intermediate training checkpoints and optimizer states for research into training dynamics
- A detailed model card on data mix, hardware, energy use and limitations
- A worked example of open, multi-institution model governance
BLOOM is a landmark of open, documented model building: training data, code, checkpoints and optimizer states were all published, and the weights are still on Hugging Face. Its language list includes Indonesian and Vietnamese but not Thai, Malay or Tagalog.
As a working model it is outdated and expensive to run, no hosted provider serves it, and its model card warns that its own evaluation results are being corrected. Use it to study multilingual training; choose a current open model for anything you ship.
What we found
BLOOM is the 176-billion-parameter language model produced by BigScience, a year-long open research workshop coordinated by Hugging Face, and released in July 2022. Hugging Face's launch post described it as the first multilingual model of that size trained in complete transparency, built by more than 1,000 researchers from over 70 countries and 250 institutions. The final run lasted 117 days on 384 A100 GPUs of France's Jean Zay supercomputer, paid for by a compute grant from the French research bodies CNRS and GENCI that the post valued at about three million euros.
The model covers 46 natural languages and 13 programming languages. For ASEAN readers the useful detail is which: Indonesian, Vietnamese and Chinese are in, Thai, Malay and Tagalog are not. Earlier versions of this page also claimed especially strong African coverage, which overstates it; the model card shows Niger-Congo languages such as Swahili and Yoruba as tiny fractions of the training text.
Everything is still downloadable. The weights, smaller sizes from 560 million to 7.1 billion parameters, intermediate checkpoints, optimizer states and the BLOOMZ instruction-tuned family all remain on Hugging Face, and the repository was last edited in July 2026, though only to fix carbon-emissions metadata; the weights themselves have not changed since 2023. No Hugging Face inference provider serves it, so running the full model means your own multi-GPU hardware.
Two cautions. The licence is the BigScience RAIL License v1.0, which permits commercial use but imposes use restrictions and is not an OSI open-source licence. And the model card carries a warning that its zero-shot evaluation results were wrong and published without the evaluation group's approval, with fixes still pending. Later open models outperform BLOOM at a fraction of the cost to run, so it now serves research into multilingual training and open science rather than production work, and that is where it still earns attention.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
Notable facts
- BLOOM was trained using the equivalent computational resources of 360 tonnes of CO2 emissions — for which BigScience purchased carbon offsets, making it one of the first 'carbon-neutral' LLMs.
- Over 1,000 researchers from 250+ institutions contributed to BLOOM's training, making it the most collaborative single AI model ever created.
- BLOOM supports languages including Swahili, Nigerian Pidgin, and Catalan that are entirely absent from most commercial LLMs.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including BLOOM's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with BLOOM unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to BLOOM directly →
Spotted something out of date? Suggest an update →
Related reading
Health & Wellness
Fine-Particle Air Pollution Linked to Higher Hospital Admission Risk for Up to 21 Days, St...
Health & Wellness
A Guide for New Mothers Missed Its Main Goal but Nearly Doubled Primary Care Visits
AI & ML
Google Stopped Paying for One Kind of Open-Source Bug Report After a Flood of Invalid Ones
Alternatives to BLOOM
More in LLMs & Chat
