Education

Galactica

Explore Galactica’s scientific language models, research access, separate code and model licenses, and the need to verify generated facts and citations.

Updated October 24, 2026

Access note: The model card and code repository remain available. We could not verify a working hosted demo at the project website; that does not establish that the research project has closed.

Image context: The existing image shows the historical Galactica demo landing page. It does not verify that a hosted demo is available today.

Field	Details
Name	Galactica
Overview	Galactica is a family of scientific language models released in November 2022 by the Papers with Code team at Meta AI. Its research paper, model files and code support experiments in scientific language modeling. The model card warns that generated text and citations can be inaccurate, even when they sound authoritative. These are research resources, not a verified current freemium web-chat service.
Key features & benefits	Research tasks including scientific question answering, summarization and document generation. Experiments with citation prediction, mathematical reasoning and scientific data. Published research paper, model cards, downloadable models and example code. Documented limitations, including fabricated information and citations. Model sizes ranging from 125 million to 120 billion parameters, with different compute requirements.
Use cases and applications	Research on language models for scientific natural language processing. Analysis of language patterns and generated scientific text. Controlled experiments with scientific question-answering tools, using safeguards and independent evaluation rather than unchecked production deployment.
Who uses?	Researchers Data scientists Developers
Pricing	Research code and model files are available under their published licenses; a current paid subscription or free web-app tier was not verified. Running a model can require your own computing or hosting resources. The model license is CC BY-NC 4.0; the repository’s code license is Apache-2.0. Check the applicable terms before reuse.
Tags	AI, Language Modeling, Research, Data Science
App available?	The documented access route is through research code and model files, including Python examples. A currently usable hosted demo or dedicated consumer app was not verified in this audit.

How to Evaluate Galactica for Research

Start with the model card and repository documentation. Choose a model size that fits your available computing resources and check the model and code licenses separately.
Test a narrow scientific task using nonconfidential examples. Keep the prompt and output so another researcher can inspect the result.
Open every cited paper and check quotations, equations and factual claims against the original sources. Do not treat a generated reference as proof that a paper exists.

The model card cautions against production use without safeguards. An unchecked response should not be treated as scientific evidence.

Primary sources: Galactica model card, official code repository, and Galactica research paper.