GPT-2 Models by OpenAI

OpenAIMITOpen source

OpenAI releases · 44 in the last 12 monthsChangelog →

This model family is considered obsolete. Consider newer alternatives in Related Model Families below.

4 models2019Up to 1K ctx

Last refreshed 2026-10-12. Next refresh: weekly.

Details

ResearcherOpenAI

LicenseMITOSI-approved

Commercial usePermitted

Models4

Released2019

Max context1K

Links

Website HuggingFace

About

GPT-2 is OpenAI’s 2019 transformer language-model family, trained to predict the next word on a large web-text corpus and released as a research successor to GPT. The largest documented model has 1.5 billion parameters and showed unsupervised language modeling plus rudimentary reading comprehension, translation, question answering, and summarization without task-specific fine-tuning. Historical checkpoints (gpt2, gpt2-medium, gpt2-large, gpt2-xl) are maintained for the community under the openai-community Hugging Face org.

Decision facts

Best fit: General model comparison
Capability starting point: GPT-2 XL with 1K context
Lowest tracked input: Not tracked
Closest related family: GPT-5

Current Variants

Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.

4 in view

GPT-2 XLCurrent

Use when the workload needs 1K context and 1.5B parameters.

2019-111K context1.5B parameters

GPT-2 LargeCurrent

Use when the workload needs 1K context and 774M parameters.

2019-081K context774M parameters

GPT-2Current

Use when the workload needs 1K context and 124M parameters.

2019-021K context124M parameters

GPT-2 MediumCurrent

Use when the workload needs 1K context and 355M parameters.

2019-021K context355M parameters

Current GPT-2 variants with use-when guidance and lifecycle status
Model	Use when	Released	Signals	Status
GPT-2 XL	Use when the workload needs 1K context and 1.5B parameters.	2019-11	1K context1.5B parameters	Current
GPT-2 Large	Use when the workload needs 1K context and 774M parameters.	2019-08	1K context774M parameters	Current
GPT-2	Use when the workload needs 1K context and 124M parameters.	2019-02	1K context124M parameters	Current
GPT-2 Medium	Use when the workload needs 1K context and 355M parameters.	2019-02	1K context355M parameters	Current