Skip to content
← Models
Open weightsMeta AI/10 trustMetrics-checked

Galactica

The large language model trained to reason about science itself.

Visit official page →

About

Galactica (November 2022) was Meta AI's attempt at a language model for science: 106B tokens of papers, textbooks, encyclopedias, and knowledge bases, trained in sizes from 125M to 120B parameters. It could do citation prediction, scientific QA, math reasoning, molecular property prediction, and protein annotation using special prompt tokens.

Its public demo was pulled after three days amid criticism over confident hallucinations — a famous cautionary tale. The weights were later released on Hugging Face under a non-commercial CC BY-NC 4.0 license. It is historically important but dated; use it for research into scientific LMs, not as a current assistant.

Specs

Context window

Not stated on the model card — decoder-only transformer in the OPT family.

Pricing

Free for non-commercial use (CC BY-NC 4.0).

Free quota

All sizes (125M–120B) on Hugging Face.

API access

Hugging Face checkpoints; galai Python package

Weights

Open weights

Released

November 1, 2022

Benchmarks

Evaluated on scientific QA, mathematical reasoning, and molecular property prediction in the release paper (arXiv 2211.09085); strong on in-distribution scientific tasks, weak at calibrated honesty — see the paper and its reception for details.

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

researchcode

Trust score — 6/10

/10 trustMetrics-checked

Scored 6/10 (metrics tier, metrics-v1). Evidence: historically significant; limited current practitioner use; effectively unmaintained since 2022; weights available under non-commercial CC BY-NC 4.0. Manual review: important cautionary tale on hallucination; dated — research interest only. License: CC BY-NC 4.0 (non-commercial)..

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
6.0/10
Maintenance · recency, releases, issue load
3.0/10
Openness · license + docs
7.0/10
Manual review · sanity read + red-flag check
6.0/10
  • · historically significant; limited current practitioner use
  • · effectively unmaintained since 2022
  • · weights available under non-commercial CC BY-NC 4.0
  • · manual review: important cautionary tale on hallucination; dated — research interest only

Sources