Tier 3 Practical open-weight
What the publisher claims: open (staged release)
The one-liner: The original open-weight model, and the original missing-dataset problem.
The six questions
All four sizes downloadable since the staged 2019 release.
Modified MIT grants use, modification, redistribution.
Runs on a laptop by today's standards.
WebText (Reddit-linked webpages) never released; only described.
Corpus unavailable; reproduction impossible in the strict sense.
openai/gpt-2 is archived; the 2019 release was one-way with no participation path.
License
Modified MIT License (adds an output-ownership paragraph; not verbatim OSI-approved MIT text) — license source ↗
- Modified MIT: OpenAI disclaims ownership of generated content
- WebText training data never released
What is actually released
- 1.5B down to 124M weights (staged release over 2019)
- code (MIT repo)
Why it matters
GPT-2 set the template every later release followed: weights yes, data no — which is exactly why the rubric scores Access and Transparency separately.
Sources
- https://raw.githubusercontent.com/openai/gpt-2/master/LICENSE
- https://huggingface.co/openai-community/gpt2
Derived from the openness dataset behind But Is It Open? — snapshot 2026-11-07 ↗. Scores are qualitative readings of the cited evidence, not verdicts; tier measures openness, not intelligence.