Introduction

This repository provides a brief summarization of the OpenAI model API.

Since the OpenAI API is updated and changes quickly, it is usually hard for researchers and developers to decide which OpenAI model to use, and what's the API name of a specific model.

To make it easier for researchers to refer to, I displayed several popular NLP models (text only; no code , embedding, image, or audio models) and attached each model with critical attributes (e.g., token limit, price, model size).

News

• 08/22/2023: Update GPT-3 based model and the new fine-tuning support of ChatGPT. See the latest announcement.
• 06/14/2023: Update the table according to OpenAI's latest announcement (06/13).
• 04/07/2023: Initial upload.

Summarization

You can also find the following table on the google doc.

📝 Some Notes:

1. The prices of Ada, Babbage, Curie, Davinci are the fine-tuning cost, while the prices of gpt-4 series are the cost of Completion. See OpenAI Pricing for more details.
2. The naming convention of API is {capability}-{family}[-{input-type}]-{identifier}. Pls refer to this website for more details.
3. The paper-report size of InstructGPT is 175B (i.e., davinci-instruct-beta), but there is no official model size information for the other APIs. Therefore, I fill in the table with my guess according to the API name (i.e., davinci). See this forum.
4. Some of the information, such as API name, might be changed over time. I will try to keep it up-to-date.

Feel free to reach out or submit a pull request for any fallacies.

References

1. OpenAI models doc

2. OpenAI models pricing

3. OpenAI Model index for researchers

4. On the Sizes of OpenAI API Models

5. How do text-davinci-002 and text-davinci-003 differ?