OpenAI Models

OpenAI is a research lab that also offers APIs and end-user products to access language models that currently push the state of the art. Their GPT-4 model is currently considered the most advanced large language model according to a variety of benchmarks.

Airtrain supports OpenAI models by calling the OpenAI API on your behalf. You do not need your own API key to call OpenAI via Airtrain.

Supported variants

gpt-4

GPT-4 was at the frontier of language modeling for several months, and still remains a powerful model despite both closed weight and open weight (ex: Llama 3 405b) competitors nearing its level of capability. It is multilingual, supports function calling, and supports working with image data as both inputs and outputs (though Airtrain only supports interactions via text).

GPT-4-turbo

A faster, cheaper variant of GPT-4 that is also often considered more capable. It was released after the original GPT-4.

gpt-4o

A faster, cheaper variant of GPT-4 that is also often considered more capable. The "o" stands for "omni," as this model is multimodal supporting text/image/audio as both inputs and outputs (Airtrain only supports text interactions). This is the primary gpt-4o model recommended for usage with the API.

chatgpt-4o-latest

The tuning of GPT-4o that is used in the ChatGPT web app, which differs from the gpt-4o model offered via the web API.

gpt-4o-mini

A smaller, faster, significantly cheaper version of gpt-4o.

o1-mini

The smaller/faster/cheaper of the two "reasoning" models trained with reinforcement learning . to excel at problem solving. It is better than the gpt models at problem solving, but considered less capable at creative writing and more open-ended tasks. Compared to o1-preview, it is significantly cheaper yet outperforms it on certain specific tasks like coding.

o1-preview

The larger/slower/more expensive of the two "reasoning" models trained with reinforcement learning . to excel at problem solving. It is better than the gpt models at problem solving, but considered less capable at creative writing and more open-ended tasks. Compared to o1-mini, it is more expensive and performs better on a broader class of problems.