Jurassic-1

What it is

Jurassic-1 is AI21 Labs' first family of large language models. A large language model is a program trained on large amounts of text to predict and generate language, and developers use such models to write, summarise, answer questions and more. AI21 announced Jurassic-1 together with the AI21 Studio developer platform in August 2021. AI21's announcement is dated 4 August 2021 ^[1][2]. The family name Jurassic is also used for the models that followed, Jurassic-2 among them.

Origins

The family was offered to developers through AI21 Studio. When Jurassic-2 was released in March 2023 it became the successor to Jurassic-1 ^[3].

Main features

The largest model, Jurassic-1 Jumbo, has 178 billion parameters, the adjustable numbers that store what a model has learned ^[1]. AI21 called it the largest and most sophisticated language model then released for general use by developers ^[1].

The model uses a vocabulary of about 250,000 tokens. A token is a unit of text, which may be a word, part of a word or a punctuation mark. AI21 said this vocabulary was at least five times larger than most existing vocabularies. It includes multi-word tokens, such as phrases and named entities, so that a common expression can be handled as one unit ^[1].

AI21 offered the models through AI21 Studio in an open beta, which anyone could sign up for. Developers could use them through an API or through an interactive web environment ^[1]. The model itself is closed: it was available only through AI21's API ^[1]. A closed model is one whose weights are not published, so outside users can use it only through the provider's service rather than running it themselves.

Place in development

In March 2023 Jurassic-1 pricing moved to a reduced model based on the total length of text, when Jurassic-2 was released as its successor ^[3]. The price change was made when Jurassic-2 arrived.