GPT-3 is 📝OpenAI's third-generation 📝GPT language model, released through a private API beta on June 11, 2020, and the first GPT model OpenAI commercialized broadly.
With 175 billion parameters, GPT-3 was more than a hundred times larger than GPT-2 and showed that a single pre-trained model could translate, summarize, answer questions, and write code from just a few examples in the prompt — a behavior OpenAI's paper called few-shot learning. It launched the first wave of startups built on a third-party model API, and in September 2020 📝Microsoft licensed it exclusively. Its instruction-tuned descendants, known as GPT-3.5, became the model behind 📝ChatGPT when it launched in November 2022.
Key Facts
- Developer: OpenAI
- Paper: "Language Models Are Few-Shot Learners," published May 28, 2020
- API launch: June 11, 2020 (private beta)
- Size: 175 billion parameters
- API variants: davinci, curie, babbage, and ada, retired January 4, 2024
- Successors: GPT-3.5 (2022) and the 📝GPT-4-series (2023)
FAQ
When was GPT-3 released?
OpenAI published the GPT-3 paper on May 28, 2020 and opened access through a private API beta on June 11, 2020.
How big is GPT-3?
GPT-3 has 175 billion parameters, more than a hundred times the 1.5 billion of GPT-2.
Is GPT-3 still available?
No. OpenAI retired the original GPT-3 API models on January 4, 2024, pointing developers to newer replacements.
Examples
Related
- 📝OpenAI Model Glossary — every OpenAI model by family