Gemini

Gemini is a multimodal LLM originating from Google's DeepMind lab. Its most powerful variant, "Ultra," is purportedly on par with OpenAI's GPT-4, placing it in the category of "best of class" models. One notable aspect of the Gemini models is their performance on long contexts and their avoidance of "position bias" (when models are better at pulling information from the beginning and end of the context than from the middle).

Airtrain supports Gemini via Google's API. You do not need your own API key to call Gemini via Airtrain.

Supported variants

Gemini Pro 1.0

Gemini Pro is a variant of Gemini whose capabilities are considered to be roughly on par with GPT-3.5.

Gemini Flash 1.5

Gemini Flash is comparable in both speed and cost to Claude Haiku. However, it is multimodal and supports a larger context window (up to 1 million tokens). In Airtrain, only textual interactions are supported.